Accelerating large-scale simulation of seismic wave propagation by multi-GPUs and three-dimensional domain decomposition

NASA Astrophysics Data System (ADS)

Okamoto, Taro; Takenaka, Hiroshi; Nakamura, Takeshi; Aoki, Takayuki

2010-12-01

We adopted the GPU (graphics processing unit) to accelerate the large-scale finite-difference simulation of seismic wave propagation. The simulation can benefit from the high-memory bandwidth of GPU because it is a "memory intensive" problem. In a single-GPU case we achieved a performance of about 56 GFlops, which was about 45-fold faster than that achieved by a single core of the host central processing unit (CPU). We confirmed that the optimized use of fast shared memory and registers were essential for performance. In the multi-GPU case with three-dimensional domain decomposition, the non-contiguous memory alignment in the ghost zones was found to impose quite long time in data transfer between GPU and the host node. This problem was solved by using contiguous memory buffers for ghost zones. We achieved a performance of about 2.2 TFlops by using 120 GPUs and 330 GB of total memory: nearly (or more than) 2200 cores of host CPUs would be required to achieve the same performance. The weak scaling was nearly proportional to the number of GPUs. We therefore conclude that GPU computing for large-scale simulation of seismic wave propagation is a promising approach as a faster simulation is possible with reduced computational resources compared to CPUs.

Benchmarking and Evaluating Unified Memory for OpenMP GPU Offloading

DOE Office of Scientific and Technical Information (OSTI.GOV)

Mishra, Alok; Li, Lingda; Kong, Martin

Here, the latest OpenMP standard offers automatic device offloading capabilities which facilitate GPU programming. Despite this, there remain many challenges. One of these is the unified memory feature introduced in recent GPUs. GPUs in current and future HPC systems have enhanced support for unified memory space. In such systems, CPU and GPU can access each other's memory transparently, that is, the data movement is managed automatically by the underlying system software and hardware. Memory over subscription is also possible in these systems. However, there is a significant lack of knowledge about how this mechanism will perform, and how programmers shouldmore » use it. We have modified several benchmarks codes, in the Rodinia benchmark suite, to study the behavior of OpenMP accelerator extensions and have used them to explore the impact of unified memory in an OpenMP context. We moreover modified the open source LLVM compiler to allow OpenMP programs to exploit unified memory. The results of our evaluation reveal that, while the performance of unified memory is comparable with that of normal GPU offloading for benchmarks with little data reuse, it suffers from significant overhead when GPU memory is over subcribed for benchmarks with large amount of data reuse. Based on these results, we provide several guidelines for programmers to achieve better performance with unified memory.« less

A GPU-Accelerated Approach for Feature Tracking in Time-Varying Imagery Datasets.

PubMed

Peng, Chao; Sahani, Sandip; Rushing, John

2017-10-01

We propose a novel parallel connected component labeling (CCL) algorithm along with efficient out-of-core data management to detect and track feature regions of large time-varying imagery datasets. Our approach contributes to the big data field with parallel algorithms tailored for GPU architectures. We remove the data dependency between frames and achieve pixel-level parallelism. Due to the large size, the entire dataset cannot fit into cached memory. Frames have to be streamed through the memory hierarchy (disk to CPU main memory and then to GPU memory), partitioned, and processed as batches, where each batch is small enough to fit into the GPU. To reconnect the feature regions that are separated due to data partitioning, we present a novel batch merging algorithm to extract the region connection information across multiple batches in a parallel fashion. The information is organized in a memory-efficient structure and supports fast indexing on the GPU. Our experiment uses a commodity workstation equipped with a single GPU. The results show that our approach can efficiently process a weather dataset composed of terabytes of time-varying radar images. The advantages of our approach are demonstrated by comparing to the performance of an efficient CPU cluster implementation which is being used by the weather scientists.

Simulation of reaction diffusion processes over biologically relevant size and time scales using multi-GPU workstations

PubMed Central

Hallock, Michael J.; Stone, John E.; Roberts, Elijah; Fry, Corey; Luthey-Schulten, Zaida

2014-01-01

Simulation of in vivo cellular processes with the reaction-diffusion master equation (RDME) is a computationally expensive task. Our previous software enabled simulation of inhomogeneous biochemical systems for small bacteria over long time scales using the MPD-RDME method on a single GPU. Simulations of larger eukaryotic systems exceed the on-board memory capacity of individual GPUs, and long time simulations of modest-sized cells such as yeast are impractical on a single GPU. We present a new multi-GPU parallel implementation of the MPD-RDME method based on a spatial decomposition approach that supports dynamic load balancing for workstations containing GPUs of varying performance and memory capacity. We take advantage of high-performance features of CUDA for peer-to-peer GPU memory transfers and evaluate the performance of our algorithms on state-of-the-art GPU devices. We present parallel e ciency and performance results for simulations using multiple GPUs as system size, particle counts, and number of reactions grow. We also demonstrate multi-GPU performance in simulations of the Min protein system in E. coli. Moreover, our multi-GPU decomposition and load balancing approach can be generalized to other lattice-based problems. PMID:24882911

Simulation of reaction diffusion processes over biologically relevant size and time scales using multi-GPU workstations.

PubMed

Hallock, Michael J; Stone, John E; Roberts, Elijah; Fry, Corey; Luthey-Schulten, Zaida

2014-05-01

Simulation of in vivo cellular processes with the reaction-diffusion master equation (RDME) is a computationally expensive task. Our previous software enabled simulation of inhomogeneous biochemical systems for small bacteria over long time scales using the MPD-RDME method on a single GPU. Simulations of larger eukaryotic systems exceed the on-board memory capacity of individual GPUs, and long time simulations of modest-sized cells such as yeast are impractical on a single GPU. We present a new multi-GPU parallel implementation of the MPD-RDME method based on a spatial decomposition approach that supports dynamic load balancing for workstations containing GPUs of varying performance and memory capacity. We take advantage of high-performance features of CUDA for peer-to-peer GPU memory transfers and evaluate the performance of our algorithms on state-of-the-art GPU devices. We present parallel e ciency and performance results for simulations using multiple GPUs as system size, particle counts, and number of reactions grow. We also demonstrate multi-GPU performance in simulations of the Min protein system in E. coli . Moreover, our multi-GPU decomposition and load balancing approach can be generalized to other lattice-based problems.

The Metropolis Monte Carlo method with CUDA enabled Graphic Processing Units

DOE Office of Scientific and Technical Information (OSTI.GOV)

Hall, Clifford; School of Physics, Astronomy, and Computational Sciences, George Mason University, 4400 University Dr., Fairfax, VA 22030; Ji, Weixiao

2014-02-01

We present a CPU–GPU system for runtime acceleration of large molecular simulations using GPU computation and memory swaps. The memory architecture of the GPU can be used both as container for simulation data stored on the graphics card and as floating-point code target, providing an effective means for the manipulation of atomistic or molecular data on the GPU. To fully take advantage of this mechanism, efficient GPU realizations of algorithms used to perform atomistic and molecular simulations are essential. Our system implements a versatile molecular engine, including inter-molecule interactions and orientational variables for performing the Metropolis Monte Carlo (MMC) algorithm,more » which is one type of Markov chain Monte Carlo. By combining memory objects with floating-point code fragments we have implemented an MMC parallel engine that entirely avoids the communication time of molecular data at runtime. Our runtime acceleration system is a forerunner of a new class of CPU–GPU algorithms exploiting memory concepts combined with threading for avoiding bus bandwidth and communication. The testbed molecular system used here is a condensed phase system of oligopyrrole chains. A benchmark shows a size scaling speedup of 60 for systems with 210,000 pyrrole monomers. Our implementation can easily be combined with MPI to connect in parallel several CPU–GPU duets. -- Highlights: •We parallelize the Metropolis Monte Carlo (MMC) algorithm on one CPU—GPU duet. •The Adaptive Tempering Monte Carlo employs MMC and profits from this CPU—GPU implementation. •Our benchmark shows a size scaling-up speedup of 62 for systems with 225,000 particles. •The testbed involves a polymeric system of oligopyrroles in the condensed phase. •The CPU—GPU parallelization includes dipole—dipole and Mie—Jones classic potentials.« less

GPU accelerated cell-based adaptive mesh refinement on unstructured quadrilateral grid

NASA Astrophysics Data System (ADS)

Luo, Xisheng; Wang, Luying; Ran, Wei; Qin, Fenghua

2016-10-01

A GPU accelerated inviscid flow solver is developed on an unstructured quadrilateral grid in the present work. For the first time, the cell-based adaptive mesh refinement (AMR) is fully implemented on GPU for the unstructured quadrilateral grid, which greatly reduces the frequency of data exchange between GPU and CPU. Specifically, the AMR is processed with atomic operations to parallelize list operations, and null memory recycling is realized to improve the efficiency of memory utilization. It is found that results obtained by GPUs agree very well with the exact or experimental results in literature. An acceleration ratio of 4 is obtained between the parallel code running on the old GPU GT9800 and the serial code running on E3-1230 V2. With the optimization of configuring a larger L1 cache and adopting Shared Memory based atomic operations on the newer GPU C2050, an acceleration ratio of 20 is achieved. The parallelized cell-based AMR processes have achieved 2x speedup on GT9800 and 18x on Tesla C2050, which demonstrates that parallel running of the cell-based AMR method on GPU is feasible and efficient. Our results also indicate that the new development of GPU architecture benefits the fluid dynamics computing significantly.

NMF-mGPU: non-negative matrix factorization on multi-GPU systems.

PubMed

Mejía-Roa, Edgardo; Tabas-Madrid, Daniel; Setoain, Javier; García, Carlos; Tirado, Francisco; Pascual-Montano, Alberto

2015-02-13

In the last few years, the Non-negative Matrix Factorization ( NMF ) technique has gained a great interest among the Bioinformatics community, since it is able to extract interpretable parts from high-dimensional datasets. However, the computing time required to process large data matrices may become impractical, even for a parallel application running on a multiprocessors cluster. In this paper, we present NMF-mGPU, an efficient and easy-to-use implementation of the NMF algorithm that takes advantage of the high computing performance delivered by Graphics-Processing Units ( GPUs ). Driven by the ever-growing demands from the video-games industry, graphics cards usually provided in PCs and laptops have evolved from simple graphics-drawing platforms into high-performance programmable systems that can be used as coprocessors for linear-algebra operations. However, these devices may have a limited amount of on-board memory, which is not considered by other NMF implementations on GPU. NMF-mGPU is based on CUDA ( Compute Unified Device Architecture ), the NVIDIA's framework for GPU computing. On devices with low memory available, large input matrices are blockwise transferred from the system's main memory to the GPU's memory, and processed accordingly. In addition, NMF-mGPU has been explicitly optimized for the different CUDA architectures. Finally, platforms with multiple GPUs can be synchronized through MPI ( Message Passing Interface ). In a four-GPU system, this implementation is about 120 times faster than a single conventional processor, and more than four times faster than a single GPU device (i.e., a super-linear speedup). Applications of GPUs in Bioinformatics are getting more and more attention due to their outstanding performance when compared to traditional processors. In addition, their relatively low price represents a highly cost-effective alternative to conventional clusters. In life sciences, this results in an excellent opportunity to facilitate the daily work of bioinformaticians that are trying to extract biological meaning out of hundreds of gigabytes of experimental information. NMF-mGPU can be used "out of the box" by researchers with little or no expertise in GPU programming in a variety of platforms, such as PCs, laptops, or high-end GPU clusters. NMF-mGPU is freely available at https://github.com/bioinfo-cnb/bionmf-gpu .

GPU-based optimal control for RWM feedback in tokamaks

DOE PAGES

Clement, Mitchell; Hanson, Jeremy; Bialek, Jim; ...

2017-08-23

The design and implementation of a Graphics Processing Unit (GPU) based Resistive Wall Mode (RWM) controller to perform feedback control on the RWM using Linear Quadratic Gaussian (LQG) control is reported herein. Also, the control algorithm is based on a simplified DIII-D VALEN model. By using NVIDIA’s GPUDirect RDMA framework, the digitizer and output module are able to write and read directly to and from GPU memory, eliminating memory transfers between host and GPU. In conclusion, the system and algorithm was able to reduce plasma response excited by externally applied fields by 32% during development experiments.

GPU-based optimal control for RWM feedback in tokamaks

DOE Office of Scientific and Technical Information (OSTI.GOV)

Clement, Mitchell; Hanson, Jeremy; Bialek, Jim

The design and implementation of a Graphics Processing Unit (GPU) based Resistive Wall Mode (RWM) controller to perform feedback control on the RWM using Linear Quadratic Gaussian (LQG) control is reported herein. Also, the control algorithm is based on a simplified DIII-D VALEN model. By using NVIDIA’s GPUDirect RDMA framework, the digitizer and output module are able to write and read directly to and from GPU memory, eliminating memory transfers between host and GPU. In conclusion, the system and algorithm was able to reduce plasma response excited by externally applied fields by 32% during development experiments.

A sample implementation for parallelizing Divide-and-Conquer algorithms on the GPU.

PubMed

Mei, Gang; Zhang, Jiayin; Xu, Nengxiong; Zhao, Kunyang

2018-01-01

The strategy of Divide-and-Conquer (D&C) is one of the frequently used programming patterns to design efficient algorithms in computer science, which has been parallelized on shared memory systems and distributed memory systems. Tzeng and Owens specifically developed a generic paradigm for parallelizing D&C algorithms on modern Graphics Processing Units (GPUs). In this paper, by following the generic paradigm proposed by Tzeng and Owens, we provide a new and publicly available GPU implementation of the famous D&C algorithm, QuickHull, to give a sample and guide for parallelizing D&C algorithms on the GPU. The experimental results demonstrate the practicality of our sample GPU implementation. Our research objective in this paper is to present a sample GPU implementation of a classical D&C algorithm to help interested readers to develop their own efficient GPU implementations with fewer efforts.

Thread scheduling for GPU-based OPC simulation on multi-thread

NASA Astrophysics Data System (ADS)

Lee, Heejun; Kim, Sangwook; Hong, Jisuk; Lee, Sooryong; Han, Hwansoo

2018-03-01

As semiconductor product development based on shrinkage continues, the accuracy and difficulty required for the model based optical proximity correction (MBOPC) is increasing. OPC simulation time, which is the most timeconsuming part of MBOPC, is rapidly increasing due to high pattern density in a layout and complex OPC model. To reduce OPC simulation time, we attempt to apply graphic processing unit (GPU) to MBOPC because OPC process is good to be programmed in parallel. We address some issues that may typically happen during GPU-based OPC simulation in multi thread system, such as "out of memory" and "GPU idle time". To overcome these problems, we propose a thread scheduling method, which manages OPC jobs in multiple threads in such a way that simulations jobs from multiple threads are alternatively executed on GPU while correction jobs are executed at the same time in each CPU cores. It was observed that the amount of GPU peak memory usage decreases by up to 35%, and MBOPC runtime also decreases by 4%. In cases where out of memory issues occur in a multi-threaded environment, the thread scheduler was used to improve MBOPC runtime up to 23%.

An Investigation of Unified Memory Access Performance in CUDA

PubMed Central

Landaverde, Raphael; Zhang, Tiansheng; Coskun, Ayse K.; Herbordt, Martin

2015-01-01

Managing memory between the CPU and GPU is a major challenge in GPU computing. A programming model, Unified Memory Access (UMA), has been recently introduced by Nvidia to simplify the complexities of memory management while claiming good overall performance. In this paper, we investigate this programming model and evaluate its performance and programming model simplifications based on our experimental results. We find that beyond on-demand data transfers to the CPU, the GPU is also able to request subsets of data it requires on demand. This feature allows UMA to outperform full data transfer methods for certain parallel applications and small data sizes. We also find, however, that for the majority of applications and memory access patterns, the performance overheads associated with UMA are significant, while the simplifications to the programming model restrict flexibility for adding future optimizations. PMID:26594668

A comparative study of history-based versus vectorized Monte Carlo methods in the GPU/CUDA environment for a simple neutron eigenvalue problem

NASA Astrophysics Data System (ADS)

Liu, Tianyu; Du, Xining; Ji, Wei; Xu, X. George; Brown, Forrest B.

2014-06-01

For nuclear reactor analysis such as the neutron eigenvalue calculations, the time consuming Monte Carlo (MC) simulations can be accelerated by using graphics processing units (GPUs). However, traditional MC methods are often history-based, and their performance on GPUs is affected significantly by the thread divergence problem. In this paper we describe the development of a newly designed event-based vectorized MC algorithm for solving the neutron eigenvalue problem. The code was implemented using NVIDIA's Compute Unified Device Architecture (CUDA), and tested on a NVIDIA Tesla M2090 GPU card. We found that although the vectorized MC algorithm greatly reduces the occurrence of thread divergence thus enhancing the warp execution efficiency, the overall simulation speed is roughly ten times slower than the history-based MC code on GPUs. Profiling results suggest that the slow speed is probably due to the memory access latency caused by the large amount of global memory transactions. Possible solutions to improve the code efficiency are discussed.

Massive parallelization of a 3D finite difference electromagnetic forward solution using domain decomposition methods on multiple CUDA enabled GPUs

NASA Astrophysics Data System (ADS)

Schultz, A.

2010-12-01

3D forward solvers lie at the core of inverse formulations used to image the variation of electrical conductivity within the Earth's interior. This property is associated with variations in temperature, composition, phase, presence of volatiles, and in specific settings, the presence of groundwater, geothermal resources, oil/gas or minerals. The high cost of 3D solutions has been a stumbling block to wider adoption of 3D methods. Parallel algorithms for modeling frequency domain 3D EM problems have not achieved wide scale adoption, with emphasis on fairly coarse grained parallelism using MPI and similar approaches. The communications bandwidth as well as the latency required to send and receive network communication packets is a limiting factor in implementing fine grained parallel strategies, inhibiting wide adoption of these algorithms. Leading Graphics Processor Unit (GPU) companies now produce GPUs with hundreds of GPU processor cores per die. The footprint, in silicon, of the GPU's restricted instruction set is much smaller than the general purpose instruction set required of a CPU. Consequently, the density of processor cores on a GPU can be much greater than on a CPU. GPUs also have local memory, registers and high speed communication with host CPUs, usually through PCIe type interconnects. The extremely low cost and high computational power of GPUs provides the EM geophysics community with an opportunity to achieve fine grained (i.e. massive) parallelization of codes on low cost hardware. The current generation of GPUs (e.g. NVidia Fermi) provides 3 billion transistors per chip die, with nearly 500 processor cores and up to 6 GB of fast (DDR5) GPU memory. This latest generation of GPU supports fast hardware double precision (64 bit) floating point operations of the type required for frequency domain EM forward solutions. Each Fermi GPU board can sustain nearly 1 TFLOP in double precision, and multiple boards can be installed in the host computer system. We describe our ongoing efforts to achieve massive parallelization on a novel hybrid GPU testbed machine currently configured with 12 Intel Westmere Xeon CPU cores (or 24 parallel computational threads) with 96 GB DDR3 system memory, 4 GPU subsystems which in aggregate contain 960 NVidia Tesla GPU cores with 16 GB dedicated DDR3 GPU memory, and a second interleved bank of 4 GPU subsystems containing in aggregate 1792 NVidia Fermi GPU cores with 12 GB dedicated DDR5 GPU memory. We are applying domain decomposition methods to a modified version of Weiss' (2001) 3D frequency domain full physics EM finite difference code, an open source GPL licensed f90 code available for download from www.OpenEM.org. This will be the core of a new hybrid 3D inversion that parallelizes frequencies across CPUs and individual forward solutions across GPUs. We describe progress made in modifying the code to use direct solvers in GPU cores dedicated to each small subdomain, iteratively improving the solution by matching adjacent subdomain boundary solutions, rather than iterative Krylov space sparse solvers as currently applied to the whole domain.

GraphReduce: Processing Large-Scale Graphs on Accelerator-Based Systems

DOE Office of Scientific and Technical Information (OSTI.GOV)

Sengupta, Dipanjan; Song, Shuaiwen; Agarwal, Kapil

2015-11-15

Recent work on real-world graph analytics has sought to leverage the massive amount of parallelism offered by GPU devices, but challenges remain due to the inherent irregularity of graph algorithms and limitations in GPU-resident memory for storing large graphs. We present GraphReduce, a highly efficient and scalable GPU-based framework that operates on graphs that exceed the device’s internal memory capacity. GraphReduce adopts a combination of edge- and vertex-centric implementations of the Gather-Apply-Scatter programming model and operates on multiple asynchronous GPU streams to fully exploit the high degrees of parallelism in GPUs with efficient graph data movement between the host andmore » device.« less

Parallelization and checkpointing of GPU applications through program transformation

DOE Office of Scientific and Technical Information (OSTI.GOV)

Solano-Quinde, Lizandro Damian

2012-01-01

GPUs have emerged as a powerful tool for accelerating general-purpose applications. The availability of programming languages that makes writing general-purpose applications for running on GPUs tractable have consolidated GPUs as an alternative for accelerating general purpose applications. Among the areas that have benefited from GPU acceleration are: signal and image processing, computational fluid dynamics, quantum chemistry, and, in general, the High Performance Computing (HPC) Industry. In order to continue to exploit higher levels of parallelism with GPUs, multi-GPU systems are gaining popularity. In this context, single-GPU applications are parallelized for running in multi-GPU systems. Furthermore, multi-GPU systems help to solvemore » the GPU memory limitation for applications with large application memory footprint. Parallelizing single-GPU applications has been approached by libraries that distribute the workload at runtime, however, they impose execution overhead and are not portable. On the other hand, on traditional CPU systems, parallelization has been approached through application transformation at pre-compile time, which enhances the application to distribute the workload at application level and does not have the issues of library-based approaches. Hence, a parallelization scheme for GPU systems based on application transformation is needed. Like any computing engine of today, reliability is also a concern in GPUs. GPUs are vulnerable to transient and permanent failures. Current checkpoint/restart techniques are not suitable for systems with GPUs. Checkpointing for GPU systems present new and interesting challenges, primarily due to the natural differences imposed by the hardware design, the memory subsystem architecture, the massive number of threads, and the limited amount of synchronization among threads. Therefore, a checkpoint/restart technique suitable for GPU systems is needed. The goal of this work is to exploit higher levels of parallelism and to develop support for application-level fault tolerance in applications using multiple GPUs. Our techniques reduce the burden of enhancing single-GPU applications to support these features. To achieve our goal, this work designs and implements a framework for enhancing a single-GPU OpenCL application through application transformation.« less

CUDA Optimization Strategies for Compute- and Memory-Bound Neuroimaging Algorithms

PubMed Central

Lee, Daren; Dinov, Ivo; Dong, Bin; Gutman, Boris; Yanovsky, Igor; Toga, Arthur W.

2011-01-01

As neuroimaging algorithms and technology continue to grow faster than CPU performance in complexity and image resolution, data-parallel computing methods will be increasingly important. The high performance, data-parallel architecture of modern graphical processing units (GPUs) can reduce computational times by orders of magnitude. However, its massively threaded architecture introduces challenges when GPU resources are exceeded. This paper presents optimization strategies for compute- and memory-bound algorithms for the CUDA architecture. For compute-bound algorithms, the registers are reduced through variable reuse via shared memory and the data throughput is increased through heavier thread workloads and maximizing the thread configuration for a single thread block per multiprocessor. For memory-bound algorithms, fitting the data into the fast but limited GPU resources is achieved through reorganizing the data into self-contained structures and employing a multi-pass approach. Memory latencies are reduced by selecting memory resources whose cache performance are optimized for the algorithm's access patterns. We demonstrate the strategies on two computationally expensive algorithms and achieve optimized GPU implementations that perform up to 6× faster than unoptimized ones. Compared to CPU implementations, we achieve peak GPU speedups of 129× for the 3D unbiased nonlinear image registration technique and 93× for the non-local means surface denoising algorithm. PMID:21159404

CUDA optimization strategies for compute- and memory-bound neuroimaging algorithms.

PubMed

Lee, Daren; Dinov, Ivo; Dong, Bin; Gutman, Boris; Yanovsky, Igor; Toga, Arthur W

2012-06-01

As neuroimaging algorithms and technology continue to grow faster than CPU performance in complexity and image resolution, data-parallel computing methods will be increasingly important. The high performance, data-parallel architecture of modern graphical processing units (GPUs) can reduce computational times by orders of magnitude. However, its massively threaded architecture introduces challenges when GPU resources are exceeded. This paper presents optimization strategies for compute- and memory-bound algorithms for the CUDA architecture. For compute-bound algorithms, the registers are reduced through variable reuse via shared memory and the data throughput is increased through heavier thread workloads and maximizing the thread configuration for a single thread block per multiprocessor. For memory-bound algorithms, fitting the data into the fast but limited GPU resources is achieved through reorganizing the data into self-contained structures and employing a multi-pass approach. Memory latencies are reduced by selecting memory resources whose cache performance are optimized for the algorithm's access patterns. We demonstrate the strategies on two computationally expensive algorithms and achieve optimized GPU implementations that perform up to 6× faster than unoptimized ones. Compared to CPU implementations, we achieve peak GPU speedups of 129× for the 3D unbiased nonlinear image registration technique and 93× for the non-local means surface denoising algorithm. Copyright © 2010 Elsevier Ireland Ltd. All rights reserved.

Multi-GPU and multi-CPU accelerated FDTD scheme for vibroacoustic applications

NASA Astrophysics Data System (ADS)

Francés, J.; Otero, B.; Bleda, S.; Gallego, S.; Neipp, C.; Márquez, A.; Beléndez, A.

2015-06-01

The Finite-Difference Time-Domain (FDTD) method is applied to the analysis of vibroacoustic problems and to study the propagation of longitudinal and transversal waves in a stratified media. The potential of the scheme and the relevance of each acceleration strategy for massively computations in FDTD are demonstrated in this work. In this paper, we propose two new specific implementations of the bi-dimensional scheme of the FDTD method using multi-CPU and multi-GPU, respectively. In the first implementation, an open source message passing interface (OMPI) has been included in order to massively exploit the resources of a biprocessor station with two Intel Xeon processors. Moreover, regarding CPU code version, the streaming SIMD extensions (SSE) and also the advanced vectorial extensions (AVX) have been included with shared memory approaches that take advantage of the multi-core platforms. On the other hand, the second implementation called the multi-GPU code version is based on Peer-to-Peer communications available in CUDA on two GPUs (NVIDIA GTX 670). Subsequently, this paper presents an accurate analysis of the influence of the different code versions including shared memory approaches, vector instructions and multi-processors (both CPU and GPU) and compares them in order to delimit the degree of improvement of using distributed solutions based on multi-CPU and multi-GPU. The performance of both approaches was analysed and it has been demonstrated that the addition of shared memory schemes to CPU computing improves substantially the performance of vector instructions enlarging the simulation sizes that use efficiently the cache memory of CPUs. In this case GPU computing is slightly twice times faster than the fine tuned CPU version in both cases one and two nodes. However, for massively computations explicit vector instructions do not worth it since the memory bandwidth is the limiting factor and the performance tends to be the same than the sequential version with auto-vectorisation and also shared memory approach. In this scenario GPU computing is the best option since it provides a homogeneous behaviour. More specifically, the speedup of GPU computing achieves an upper limit of 12 for both one and two GPUs, whereas the performance reaches peak values of 80 GFlops and 146 GFlops for the performance for one GPU and two GPUs respectively. Finally, the method is applied to an earth crust profile in order to demonstrate the potential of our approach and the necessity of applying acceleration strategies in these type of applications.

Efficient methods for implementation of multi-level nonrigid mass-preserving image registration on GPUs and multi-threaded CPUs.

PubMed

Ellingwood, Nathan D; Yin, Youbing; Smith, Matthew; Lin, Ching-Long

2016-04-01

Faster and more accurate methods for registration of images are important for research involved in conducting population-based studies that utilize medical imaging, as well as improvements for use in clinical applications. We present a novel computation- and memory-efficient multi-level method on graphics processing units (GPU) for performing registration of two computed tomography (CT) volumetric lung images. We developed a computation- and memory-efficient Diffeomorphic Multi-level B-Spline Transform Composite (DMTC) method to implement nonrigid mass-preserving registration of two CT lung images on GPU. The framework consists of a hierarchy of B-Spline control grids of increasing resolution. A similarity criterion known as the sum of squared tissue volume difference (SSTVD) was adopted to preserve lung tissue mass. The use of SSTVD consists of the calculation of the tissue volume, the Jacobian, and their derivatives, which makes its implementation on GPU challenging due to memory constraints. The use of the DMTC method enabled reduced computation and memory storage of variables with minimal communication between GPU and Central Processing Unit (CPU) due to ability to pre-compute values. The method was assessed on six healthy human subjects. Resultant GPU-generated displacement fields were compared against the previously validated CPU counterpart fields, showing good agreement with an average normalized root mean square error (nRMS) of 0.044±0.015. Runtime and performance speedup are compared between single-threaded CPU, multi-threaded CPU, and GPU algorithms. Best performance speedup occurs at the highest resolution in the GPU implementation for the SSTVD cost and cost gradient computations, with a speedup of 112 times that of the single-threaded CPU version and 11 times over the twelve-threaded version when considering average time per iteration using a Nvidia Tesla K20X GPU. The proposed GPU-based DMTC method outperforms its multi-threaded CPU version in terms of runtime. Total registration time reduced runtime to 2.9min on the GPU version, compared to 12.8min on twelve-threaded CPU version and 112.5min on a single-threaded CPU. Furthermore, the GPU implementation discussed in this work can be adapted for use of other cost functions that require calculation of the first derivatives. Copyright © 2015 Elsevier Ireland Ltd. All rights reserved.

Multi-phase SPH modelling of violent hydrodynamics on GPUs

NASA Astrophysics Data System (ADS)

Mokos, Athanasios; Rogers, Benedict D.; Stansby, Peter K.; Domínguez, José M.

2015-11-01

This paper presents the acceleration of multi-phase smoothed particle hydrodynamics (SPH) using a graphics processing unit (GPU) enabling large numbers of particles (10-20 million) to be simulated on just a single GPU card. With novel hardware architectures such as a GPU, the optimum approach to implement a multi-phase scheme presents some new challenges. Many more particles must be included in the calculation and there are very different speeds of sound in each phase with the largest speed of sound determining the time step. This requires efficient computation. To take full advantage of the hardware acceleration provided by a single GPU for a multi-phase simulation, four different algorithms are investigated: conditional statements, binary operators, separate particle lists and an intermediate global function. Runtime results show that the optimum approach needs to employ separate cell and neighbour lists for each phase. The profiler shows that this approach leads to a reduction in both memory transactions and arithmetic operations giving significant runtime gains. The four different algorithms are compared to the efficiency of the optimised single-phase GPU code, DualSPHysics, for 2-D and 3-D simulations which indicate that the multi-phase functionality has a significant computational overhead. A comparison with an optimised CPU code shows a speed up of an order of magnitude over an OpenMP simulation with 8 threads and two orders of magnitude over a single thread simulation. A demonstration of the multi-phase SPH GPU code is provided by a 3-D dam break case impacting an obstacle. This shows better agreement with experimental results than an equivalent single-phase code. The multi-phase GPU code enables a convergence study to be undertaken on a single GPU with a large number of particles that otherwise would have required large high performance computing resources.

SU-E-T-395: Multi-GPU-Based VMAT Treatment Plan Optimization Using a Column-Generation Approach

DOE Office of Scientific and Technical Information (OSTI.GOV)

Tian, Z; Shi, F; Jia, X

Purpose: GPU has been employed to speed up VMAT optimizations from hours to minutes. However, its limited memory capacity makes it difficult to handle cases with a huge dose-deposition-coefficient (DDC) matrix, e.g. those with a large target size, multiple arcs, small beam angle intervals and/or small beamlet size. We propose multi-GPU-based VMAT optimization to solve this memory issue to make GPU-based VMAT more practical for clinical use. Methods: Our column-generation-based method generates apertures sequentially by iteratively searching for an optimal feasible aperture (referred as pricing problem, PP) and optimizing aperture intensities (referred as master problem, MP). The PP requires accessmore » to the large DDC matrix, which is implemented on a multi-GPU system. Each GPU stores a DDC sub-matrix corresponding to one fraction of beam angles and is only responsible for calculation related to those angles. Broadcast and parallel reduction schemes are adopted for inter-GPU data transfer. MP is a relatively small-scale problem and is implemented on one GPU. One headand- neck cancer case was used for test. Three different strategies for VMAT optimization on single GPU were also implemented for comparison: (S1) truncating DDC matrix to ignore its small value entries for optimization; (S2) transferring DDC matrix part by part to GPU during optimizations whenever needed; (S3) moving DDC matrix related calculation onto CPU. Results: Our multi-GPU-based implementation reaches a good plan within 1 minute. Although S1 was 10 seconds faster than our method, the obtained plan quality is worse. Both S2 and S3 handle the full DDC matrix and hence yield the same plan as in our method. However, the computation time is longer, namely 4 minutes and 30 minutes, respectively. Conclusion: Our multi-GPU-based VMAT optimization can effectively solve the limited memory issue with good plan quality and high efficiency, making GPUbased ultra-fast VMAT planning practical for real clinical use.« less

GraphReduce: Large-Scale Graph Analytics on Accelerator-Based HPC Systems

DOE Office of Scientific and Technical Information (OSTI.GOV)

Sengupta, Dipanjan; Agarwal, Kapil; Song, Shuaiwen

2015-09-30

Recent work on real-world graph analytics has sought to leverage the massive amount of parallelism offered by GPU devices, but challenges remain due to the inherent irregularity of graph algorithms and limitations in GPU-resident memory for storing large graphs. We present GraphReduce, a highly efficient and scalable GPU-based framework that operates on graphs that exceed the device’s internal memory capacity. GraphReduce adopts a combination of both edge- and vertex-centric implementations of the Gather-Apply-Scatter programming model and operates on multiple asynchronous GPU streams to fully exploit the high degrees of parallelism in GPUs with efficient graph data movement between the hostmore » and the device.« less

Parallel Computational Protein Design.

PubMed

Zhou, Yichao; Donald, Bruce R; Zeng, Jianyang

2017-01-01

Computational structure-based protein design (CSPD) is an important problem in computational biology, which aims to design or improve a prescribed protein function based on a protein structure template. It provides a practical tool for real-world protein engineering applications. A popular CSPD method that guarantees to find the global minimum energy solution (GMEC) is to combine both dead-end elimination (DEE) and A* tree search algorithms. However, in this framework, the A* search algorithm can run in exponential time in the worst case, which may become the computation bottleneck of large-scale computational protein design process. To address this issue, we extend and add a new module to the OSPREY program that was previously developed in the Donald lab (Gainza et al., Methods Enzymol 523:87, 2013) to implement a GPU-based massively parallel A* algorithm for improving protein design pipeline. By exploiting the modern GPU computational framework and optimizing the computation of the heuristic function for A* search, our new program, called gOSPREY, can provide up to four orders of magnitude speedups in large protein design cases with a small memory overhead comparing to the traditional A* search algorithm implementation, while still guaranteeing the optimality. In addition, gOSPREY can be configured to run in a bounded-memory mode to tackle the problems in which the conformation space is too large and the global optimal solution cannot be computed previously. Furthermore, the GPU-based A* algorithm implemented in the gOSPREY program can be combined with the state-of-the-art rotamer pruning algorithms such as iMinDEE (Gainza et al., PLoS Comput Biol 8:e1002335, 2012) and DEEPer (Hallen et al., Proteins 81:18-39, 2013) to also consider continuous backbone and side-chain flexibility.

Power and Performance Trade-offs for Space Time Adaptive Processing

DOE Office of Scientific and Technical Information (OSTI.GOV)

Gawande, Nitin A.; Manzano Franco, Joseph B.; Tumeo, Antonino

Computational efficiency – performance relative to power or energy – is one of the most important concerns when designing RADAR processing systems. This paper analyzes power and performance trade-offs for a typical Space Time Adaptive Processing (STAP) application. We study STAP implementations for CUDA and OpenMP on two computationally efficient architectures, Intel Haswell Core I7-4770TE and NVIDIA Kayla with a GK208 GPU. We analyze the power and performance of STAP’s computationally intensive kernels across the two hardware testbeds. We also show the impact and trade-offs of GPU optimization techniques. We show that data parallelism can be exploited for efficient implementationmore » on the Haswell CPU architecture. The GPU architecture is able to process large size data sets without increase in power requirement. The use of shared memory has a significant impact on the power requirement for the GPU. A balance between the use of shared memory and main memory access leads to an improved performance in a typical STAP application.« less

Methods for compressible fluid simulation on GPUs using high-order finite differences

NASA Astrophysics Data System (ADS)

Pekkilä, Johannes; Väisälä, Miikka S.; Käpylä, Maarit J.; Käpylä, Petri J.; Anjum, Omer

2017-08-01

We focus on implementing and optimizing a sixth-order finite-difference solver for simulating compressible fluids on a GPU using third-order Runge-Kutta integration. Since graphics processing units perform well in data-parallel tasks, this makes them an attractive platform for fluid simulation. However, high-order stencil computation is memory-intensive with respect to both main memory and the caches of the GPU. We present two approaches for simulating compressible fluids using 55-point and 19-point stencils. We seek to reduce the requirements for memory bandwidth and cache size in our methods by using cache blocking and decomposing a latency-bound kernel into several bandwidth-bound kernels. Our fastest implementation is bandwidth-bound and integrates 343 million grid points per second on a Tesla K40t GPU, achieving a 3 . 6 × speedup over a comparable hydrodynamics solver benchmarked on two Intel Xeon E5-2690v3 processors. Our alternative GPU implementation is latency-bound and achieves the rate of 168 million updates per second.

SU-F-BRF-01: A GPU Framework for Developing Interactive High-Resolution Patient-Specific Biomechanical Models

DOE Office of Scientific and Technical Information (OSTI.GOV)

Neylon, J; Qi, S; Sheng, K

2014-06-15

Purpose: To develop a GPU-based framework that can generate highresolution and patient-specific biomechanical models from a given simulation CT and contoured structures, optimized to run at interactive speeds, for addressing adaptive radiotherapy objectives. Method: A Massspring-damping (MSD) model was generated from a given simulation CT. The model's mass elements were generated for every voxel of anatomy, and positioned in a deformation space in the GPU memory. MSD connections were established between neighboring mass elements in a dense distribution. Contoured internal structures allowed control over elastic material properties of different tissues. Once the model was initialized in GPU memory, skeletal anatomymore » was actuated using rigid-body transformations, while soft tissues were governed by elastic corrective forces and constraints, which included tensile forces, shear forces, and spring damping forces. The model was validated by applying a known load to a soft tissue block and comparing the observed deformation to ground truth calculations from established elastic mechanics. Results: Our analyses showed that both local and global load experiments yielded results with a correlation coefficient R{sup 2} > 0.98 compared to ground truth. Models were generated for several anatomical regions. Head and neck models accurately simulated posture changes by rotating the skeletal anatomy in three dimensions. Pelvic models were developed for realistic deformations for changes in bladder volume. Thoracic models demonstrated breast deformation due to gravity when changing treatment position from supine to prone. The GPU framework performed at greater than 30 iterations per second for over 1 million mass elements with up to 26 MSD connections each. Conclusions: Realistic simulations of site-specific, complex posture and physiological changes were simulated at interactive speeds using patient data. Incorporating such a model with live patient tracking would facilitate real time assessment of variations of the actual anatomy and delivered dose for adaptive intervention and re-planning.« less

Aspects of GPU perfomance in algorithms with random memory access

NASA Astrophysics Data System (ADS)

Kashkovsky, Alexander V.; Shershnev, Anton A.; Vashchenkov, Pavel V.

2017-10-01

The numerical code for solving the Boltzmann equation on the hybrid computational cluster using the Direct Simulation Monte Carlo (DSMC) method showed that on Tesla K40 accelerators computational performance drops dramatically with increase of percentage of occupied GPU memory. Testing revealed that memory access time increases tens of times after certain critical percentage of memory is occupied. Moreover, it seems to be the common problem of all NVidia's GPUs arising from its architecture. Few modifications of the numerical algorithm were suggested to overcome this problem. One of them, based on the splitting the memory into "virtual" blocks, resulted in 2.5 times speed up.

LU Factorization with Partial Pivoting for a Multi-CPU, Multi-GPU Shared Memory System

DOE Office of Scientific and Technical Information (OSTI.GOV)

Kurzak, Jakub; Luszczek, Pitior; Faverge, Mathieu

2012-03-01

LU factorization with partial pivoting is a canonical numerical procedure and the main component of the High Performance LINPACK benchmark. This article presents an implementation of the algorithm for a hybrid, shared memory, system with standard CPU cores and GPU accelerators. Performance in excess of one TeraFLOPS is achieved using four AMD Magny Cours CPUs and four NVIDIA Fermi GPUs.

Fast distributed large-pixel-count hologram computation using a GPU cluster.

PubMed

Pan, Yuechao; Xu, Xuewu; Liang, Xinan

2013-09-10

Large-pixel-count holograms are one essential part for big size holographic three-dimensional (3D) display, but the generation of such holograms is computationally demanding. In order to address this issue, we have built a graphics processing unit (GPU) cluster with 32.5 Tflop/s computing power and implemented distributed hologram computation on it with speed improvement techniques, such as shared memory on GPU, GPU level adaptive load balancing, and node level load distribution. Using these speed improvement techniques on the GPU cluster, we have achieved 71.4 times computation speed increase for 186M-pixel holograms. Furthermore, we have used the approaches of diffraction limits and subdivision of holograms to overcome the GPU memory limit in computing large-pixel-count holograms. 745M-pixel and 1.80G-pixel holograms were computed in 343 and 3326 s, respectively, for more than 2 million object points with RGB colors. Color 3D objects with 1.02M points were successfully reconstructed from 186M-pixel hologram computed in 8.82 s with all the above three speed improvement techniques. It is shown that distributed hologram computation using a GPU cluster is a promising approach to increase the computation speed of large-pixel-count holograms for large size holographic display.

Performance and scalability of Fourier domain optical coherence tomography acceleration using graphics processing units.

PubMed

Li, Jian; Bloch, Pavel; Xu, Jing; Sarunic, Marinko V; Shannon, Lesley

2011-05-01

Fourier domain optical coherence tomography (FD-OCT) provides faster line rates, better resolution, and higher sensitivity for noninvasive, in vivo biomedical imaging compared to traditional time domain OCT (TD-OCT). However, because the signal processing for FD-OCT is computationally intensive, real-time FD-OCT applications demand powerful computing platforms to deliver acceptable performance. Graphics processing units (GPUs) have been used as coprocessors to accelerate FD-OCT by leveraging their relatively simple programming model to exploit thread-level parallelism. Unfortunately, GPUs do not "share" memory with their host processors, requiring additional data transfers between the GPU and CPU. In this paper, we implement a complete FD-OCT accelerator on a consumer grade GPU/CPU platform. Our data acquisition system uses spectrometer-based detection and a dual-arm interferometer topology with numerical dispersion compensation for retinal imaging. We demonstrate that the maximum line rate is dictated by the memory transfer time and not the processing time due to the GPU platform's memory model. Finally, we discuss how the performance trends of GPU-based accelerators compare to the expected future requirements of FD-OCT data rates.

An efficient tensor transpose algorithm for multicore CPU, Intel Xeon Phi, and NVidia Tesla GPU

NASA Astrophysics Data System (ADS)

Lyakh, Dmitry I.

2015-04-01

An efficient parallel tensor transpose algorithm is suggested for shared-memory computing units, namely, multicore CPU, Intel Xeon Phi, and NVidia GPU. The algorithm operates on dense tensors (multidimensional arrays) and is based on the optimization of cache utilization on x86 CPU and the use of shared memory on NVidia GPU. From the applied side, the ultimate goal is to minimize the overhead encountered in the transformation of tensor contractions into matrix multiplications in computer implementations of advanced methods of quantum many-body theory (e.g., in electronic structure theory and nuclear physics). A particular accent is made on higher-dimensional tensors that typically appear in the so-called multireference correlated methods of electronic structure theory. Depending on tensor dimensionality, the presented optimized algorithms can achieve an order of magnitude speedup on x86 CPUs and 2-3 times speedup on NVidia Tesla K20X GPU with respect to the naïve scattering algorithm (no memory access optimization). The tensor transpose routines developed in this work have been incorporated into a general-purpose tensor algebra library (TAL-SH).

Fast GPU-based computation of spatial multigrid multiframe LMEM for PET.

PubMed

Nassiri, Moulay Ali; Carrier, Jean-François; Després, Philippe

2015-09-01

Significant efforts were invested during the last decade to accelerate PET list-mode reconstructions, notably with GPU devices. However, the computation time per event is still relatively long, and the list-mode efficiency on the GPU is well below the histogram-mode efficiency. Since list-mode data are not arranged in any regular pattern, costly accesses to the GPU global memory can hardly be optimized and geometrical symmetries cannot be used. To overcome obstacles that limit the acceleration of reconstruction from list-mode on the GPU, a multigrid and multiframe approach of an expectation-maximization algorithm was developed. The reconstruction process is started during data acquisition, and calculations are executed concurrently on the GPU and the CPU, while the system matrix is computed on-the-fly. A new convergence criterion also was introduced, which is computationally more efficient on the GPU. The implementation was tested on a Tesla C2050 GPU device for a Gemini GXL PET system geometry. The results show that the proposed algorithm (multigrid and multiframe list-mode expectation-maximization, MGMF-LMEM) converges to the same solution as the LMEM algorithm more than three times faster. The execution time of the MGMF-LMEM algorithm was 1.1 s per million of events on the Tesla C2050 hardware used, for a reconstructed space of 188 x 188 x 57 voxels of 2 x 2 x 3.15 mm3. For 17- and 22-mm simulated hot lesions, the MGMF-LMEM algorithm led on the first iteration to contrast recovery coefficients (CRC) of more than 75 % of the maximum CRC while achieving a minimum in the relative mean square error. Therefore, the MGMF-LMEM algorithm can be used as a one-pass method to perform real-time reconstructions for low-count acquisitions, as in list-mode gated studies. The computation time for one iteration and 60 millions of events was approximately 66 s.

Boosting the FM-Index on the GPU: Effective Techniques to Mitigate Random Memory Access.

PubMed

Chacón, Alejandro; Marco-Sola, Santiago; Espinosa, Antonio; Ribeca, Paolo; Moure, Juan Carlos

2015-01-01

The recent advent of high-throughput sequencing machines producing big amounts of short reads has boosted the interest in efficient string searching techniques. As of today, many mainstream sequence alignment software tools rely on a special data structure, called the FM-index, which allows for fast exact searches in large genomic references. However, such searches translate into a pseudo-random memory access pattern, thus making memory access the limiting factor of all computation-efficient implementations, both on CPUs and GPUs. Here, we show that several strategies can be put in place to remove the memory bottleneck on the GPU: more compact indexes can be implemented by having more threads work cooperatively on larger memory blocks, and a k-step FM-index can be used to further reduce the number of memory accesses. The combination of those and other optimisations yields an implementation that is able to process about two Gbases of queries per second on our test platform, being about 8 × faster than a comparable multi-core CPU version, and about 3 × to 5 × faster than the FM-index implementation on the GPU provided by the recently announced Nvidia NVBIO bioinformatics library.

Gpu Implementation of a Viscous Flow Solver on Unstructured Grids

NASA Astrophysics Data System (ADS)

Xu, Tianhao; Chen, Long

2016-06-01

Graphics processing units have gained popularities in scientific computing over past several years due to their outstanding parallel computing capability. Computational fluid dynamics applications involve large amounts of calculations, therefore a latest GPU card is preferable of which the peak computing performance and memory bandwidth are much better than a contemporary high-end CPU. We herein focus on the detailed implementation of our GPU targeting Reynolds-averaged Navier-Stokes equations solver based on finite-volume method. The solver employs a vertex-centered scheme on unstructured grids for the sake of being capable of handling complex topologies. Multiple optimizations are carried out to improve the memory accessing performance and kernel utilization. Both steady and unsteady flow simulation cases are carried out using explicit Runge-Kutta scheme. The solver with GPU acceleration in this paper is demonstrated to have competitive advantages over the CPU targeting one.

Accelerating three-dimensional FDTD calculations on GPU clusters for electromagnetic field simulation.

PubMed

Nagaoka, Tomoaki; Watanabe, Soichi

2012-01-01

Electromagnetic simulation with anatomically realistic computational human model using the finite-difference time domain (FDTD) method has recently been performed in a number of fields in biomedical engineering. To improve the method's calculation speed and realize large-scale computing with the computational human model, we adapt three-dimensional FDTD code to a multi-GPU cluster environment with Compute Unified Device Architecture and Message Passing Interface. Our multi-GPU cluster system consists of three nodes. The seven GPU boards (NVIDIA Tesla C2070) are mounted on each node. We examined the performance of the FDTD calculation on multi-GPU cluster environment. We confirmed that the FDTD calculation on the multi-GPU clusters is faster than that on a multi-GPU (a single workstation), and we also found that the GPU cluster system calculate faster than a vector supercomputer. In addition, our GPU cluster system allowed us to perform the large-scale FDTD calculation because were able to use GPU memory of over 100 GB.

Convolution of large 3D images on GPU and its decomposition

NASA Astrophysics Data System (ADS)

Karas, Pavel; Svoboda, David

2011-12-01

In this article, we propose a method for computing convolution of large 3D images. The convolution is performed in a frequency domain using a convolution theorem. The algorithm is accelerated on a graphic card by means of the CUDA parallel computing model. Convolution is decomposed in a frequency domain using the decimation in frequency algorithm. We pay attention to keeping our approach efficient in terms of both time and memory consumption and also in terms of memory transfers between CPU and GPU which have a significant inuence on overall computational time. We also study the implementation on multiple GPUs and compare the results between the multi-GPU and multi-CPU implementations.

GPU-Acceleration of Sequence Homology Searches with Database Subsequence Clustering.

PubMed

Suzuki, Shuji; Kakuta, Masanori; Ishida, Takashi; Akiyama, Yutaka

2016-01-01

Sequence homology searches are used in various fields and require large amounts of computation time, especially for metagenomic analysis, owing to the large number of queries and the database size. To accelerate computing analyses, graphics processing units (GPUs) are widely used as a low-cost, high-performance computing platform. Therefore, we mapped the time-consuming steps involved in GHOSTZ, which is a state-of-the-art homology search algorithm for protein sequences, onto a GPU and implemented it as GHOSTZ-GPU. In addition, we optimized memory access for GPU calculations and for communication between the CPU and GPU. As per results of the evaluation test involving metagenomic data, GHOSTZ-GPU with 12 CPU threads and 1 GPU was approximately 3.0- to 4.1-fold faster than GHOSTZ with 12 CPU threads. Moreover, GHOSTZ-GPU with 12 CPU threads and 3 GPUs was approximately 5.8- to 7.7-fold faster than GHOSTZ with 12 CPU threads.

An efficient tensor transpose algorithm for multicore CPU, Intel Xeon Phi, and NVidia Tesla GPU

DOE PAGES

Lyakh, Dmitry I.

2015-01-05

An efficient parallel tensor transpose algorithm is suggested for shared-memory computing units, namely, multicore CPU, Intel Xeon Phi, and NVidia GPU. The algorithm operates on dense tensors (multidimensional arrays) and is based on the optimization of cache utilization on x86 CPU and the use of shared memory on NVidia GPU. From the applied side, the ultimate goal is to minimize the overhead encountered in the transformation of tensor contractions into matrix multiplications in computer implementations of advanced methods of quantum many-body theory (e.g., in electronic structure theory and nuclear physics). A particular accent is made on higher-dimensional tensors that typicallymore » appear in the so-called multireference correlated methods of electronic structure theory. Depending on tensor dimensionality, the presented optimized algorithms can achieve an order of magnitude speedup on x86 CPUs and 2-3 times speedup on NVidia Tesla K20X GPU with respect to the na ve scattering algorithm (no memory access optimization). Furthermore, the tensor transpose routines developed in this work have been incorporated into a general-purpose tensor algebra library (TAL-SH).« less

A CFD Heterogeneous Parallel Solver Based on Collaborating CPU and GPU

NASA Astrophysics Data System (ADS)

Lai, Jianqi; Tian, Zhengyu; Li, Hua; Pan, Sha

2018-03-01

Since Graphic Processing Unit (GPU) has a strong ability of floating-point computation and memory bandwidth for data parallelism, it has been widely used in the areas of common computing such as molecular dynamics (MD), computational fluid dynamics (CFD) and so on. The emergence of compute unified device architecture (CUDA), which reduces the complexity of compiling program, brings the great opportunities to CFD. There are three different modes for parallel solution of NS equations: parallel solver based on CPU, parallel solver based on GPU and heterogeneous parallel solver based on collaborating CPU and GPU. As we can see, GPUs are relatively rich in compute capacity but poor in memory capacity and the CPUs do the opposite. We need to make full use of the GPUs and CPUs, so a CFD heterogeneous parallel solver based on collaborating CPU and GPU has been established. Three cases are presented to analyse the solver’s computational accuracy and heterogeneous parallel efficiency. The numerical results agree well with experiment results, which demonstrate that the heterogeneous parallel solver has high computational precision. The speedup on a single GPU is more than 40 for laminar flow, it decreases for turbulent flow, but it still can reach more than 20. What’s more, the speedup increases as the grid size becomes larger.

Lossless data compression for improving the performance of a GPU-based beamformer.

PubMed

Lok, U-Wai; Fan, Gang-Wei; Li, Pai-Chi

2015-04-01

The powerful parallel computation ability of a graphics processing unit (GPU) makes it feasible to perform dynamic receive beamforming However, a real time GPU-based beamformer requires high data rate to transfer radio-frequency (RF) data from hardware to software memory, as well as from central processing unit (CPU) to GPU memory. There are data compression methods (e.g. Joint Photographic Experts Group (JPEG)) available for the hardware front end to reduce data size, alleviating the data transfer requirement of the hardware interface. Nevertheless, the required decoding time may even be larger than the transmission time of its original data, in turn degrading the overall performance of the GPU-based beamformer. This article proposes and implements a lossless compression-decompression algorithm, which enables in parallel compression and decompression of data. By this means, the data transfer requirement of hardware interface and the transmission time of CPU to GPU data transfers are reduced, without sacrificing image quality. In simulation results, the compression ratio reached around 1.7. The encoder design of our lossless compression approach requires low hardware resources and reasonable latency in a field programmable gate array. In addition, the transmission time of transferring data from CPU to GPU with the parallel decoding process improved by threefold, as compared with transferring original uncompressed data. These results show that our proposed lossless compression plus parallel decoder approach not only mitigate the transmission bandwidth requirement to transfer data from hardware front end to software system but also reduce the transmission time for CPU to GPU data transfer. © The Author(s) 2014.

gpuPOM: a GPU-based Princeton Ocean Model

NASA Astrophysics Data System (ADS)

Xu, S.; Huang, X.; Zhang, Y.; Fu, H.; Oey, L.-Y.; Xu, F.; Yang, G.

2014-11-01

Rapid advances in the performance of the graphics processing unit (GPU) have made the GPU a compelling solution for a series of scientific applications. However, most existing GPU acceleration works for climate models are doing partial code porting for certain hot spots, and can only achieve limited speedup for the entire model. In this work, we take the mpiPOM (a parallel version of the Princeton Ocean Model) as our starting point, design and implement a GPU-based Princeton Ocean Model. By carefully considering the architectural features of the state-of-the-art GPU devices, we rewrite the full mpiPOM model from the original Fortran version into a new Compute Unified Device Architecture C (CUDA-C) version. We take several accelerating methods to further improve the performance of gpuPOM, including optimizing memory access in a single GPU, overlapping communication and boundary operations among multiple GPUs, and overlapping input/output (I/O) between the hybrid Central Processing Unit (CPU) and the GPU. Our experimental results indicate that the performance of the gpuPOM on a workstation containing 4 GPUs is comparable to a powerful cluster with 408 CPU cores and it reduces the energy consumption by 6.8 times.

GPU-Acceleration of Sequence Homology Searches with Database Subsequence Clustering

PubMed Central

Suzuki, Shuji; Kakuta, Masanori; Ishida, Takashi; Akiyama, Yutaka

2016-01-01

Sequence homology searches are used in various fields and require large amounts of computation time, especially for metagenomic analysis, owing to the large number of queries and the database size. To accelerate computing analyses, graphics processing units (GPUs) are widely used as a low-cost, high-performance computing platform. Therefore, we mapped the time-consuming steps involved in GHOSTZ, which is a state-of-the-art homology search algorithm for protein sequences, onto a GPU and implemented it as GHOSTZ-GPU. In addition, we optimized memory access for GPU calculations and for communication between the CPU and GPU. As per results of the evaluation test involving metagenomic data, GHOSTZ-GPU with 12 CPU threads and 1 GPU was approximately 3.0- to 4.1-fold faster than GHOSTZ with 12 CPU threads. Moreover, GHOSTZ-GPU with 12 CPU threads and 3 GPUs was approximately 5.8- to 7.7-fold faster than GHOSTZ with 12 CPU threads. PMID:27482905

Employing multi-GPU power for molecular dynamics simulation: an extension of GALAMOST

NASA Astrophysics Data System (ADS)

Zhu, You-Liang; Pan, Deng; Li, Zhan-Wei; Liu, Hong; Qian, Hu-Jun; Zhao, Yang; Lu, Zhong-Yuan; Sun, Zhao-Yan

2018-04-01

We describe the algorithm of employing multi-GPU power on the basis of Message Passing Interface (MPI) domain decomposition in a molecular dynamics code, GALAMOST, which is designed for the coarse-grained simulation of soft matters. The code of multi-GPU version is developed based on our previous single-GPU version. In multi-GPU runs, one GPU takes charge of one domain and runs single-GPU code path. The communication between neighbouring domains takes a similar algorithm of CPU-based code of LAMMPS, but is optimised specifically for GPUs. We employ a memory-saving design which can enlarge maximum system size at the same device condition. An optimisation algorithm is employed to prolong the update period of neighbour list. We demonstrate good performance of multi-GPU runs on the simulation of Lennard-Jones liquid, dissipative particle dynamics liquid, polymer and nanoparticle composite, and two-patch particles on workstation. A good scaling of many nodes on cluster for two-patch particles is presented.

Exploiting graphics processing units for computational biology and bioinformatics.

PubMed

Payne, Joshua L; Sinnott-Armstrong, Nicholas A; Moore, Jason H

2010-09-01

Advances in the video gaming industry have led to the production of low-cost, high-performance graphics processing units (GPUs) that possess more memory bandwidth and computational capability than central processing units (CPUs), the standard workhorses of scientific computing. With the recent release of generalpurpose GPUs and NVIDIA's GPU programming language, CUDA, graphics engines are being adopted widely in scientific computing applications, particularly in the fields of computational biology and bioinformatics. The goal of this article is to concisely present an introduction to GPU hardware and programming, aimed at the computational biologist or bioinformaticist. To this end, we discuss the primary differences between GPU and CPU architecture, introduce the basics of the CUDA programming language, and discuss important CUDA programming practices, such as the proper use of coalesced reads, data types, and memory hierarchies. We highlight each of these topics in the context of computing the all-pairs distance between instances in a dataset, a common procedure in numerous disciplines of scientific computing. We conclude with a runtime analysis of the GPU and CPU implementations of the all-pairs distance calculation. We show our final GPU implementation to outperform the CPU implementation by a factor of 1700.

Solving global shallow water equations on heterogeneous supercomputers

PubMed Central

Fu, Haohuan; Gan, Lin; Yang, Chao; Xue, Wei; Wang, Lanning; Wang, Xinliang; Huang, Xiaomeng; Yang, Guangwen

2017-01-01

The scientific demand for more accurate modeling of the climate system calls for more computing power to support higher resolutions, inclusion of more component models, more complicated physics schemes, and larger ensembles. As the recent improvements in computing power mostly come from the increasing number of nodes in a system and the integration of heterogeneous accelerators, how to scale the computing problems onto more nodes and various kinds of accelerators has become a challenge for the model development. This paper describes our efforts on developing a highly scalable framework for performing global atmospheric modeling on heterogeneous supercomputers equipped with various accelerators, such as GPU (Graphic Processing Unit), MIC (Many Integrated Core), and FPGA (Field Programmable Gate Arrays) cards. We propose a generalized partition scheme of the problem domain, so as to keep a balanced utilization of both CPU resources and accelerator resources. With optimizations on both computing and memory access patterns, we manage to achieve around 8 to 20 times speedup when comparing one hybrid GPU or MIC node with one CPU node with 12 cores. Using a customized FPGA-based data-flow engines, we see the potential to gain another 5 to 8 times improvement on performance. On heterogeneous supercomputers, such as Tianhe-1A and Tianhe-2, our framework is capable of achieving ideally linear scaling efficiency, and sustained double-precision performances of 581 Tflops on Tianhe-1A (using 3750 nodes) and 3.74 Pflops on Tianhe-2 (using 8644 nodes). Our study also provides an evaluation on the programming paradigm of various accelerator architectures (GPU, MIC, FPGA) for performing global atmospheric simulation, to form a picture about both the potential performance benefits and the programming efforts involved. PMID:28282428

Towards robust algorithms for current deposition and dynamic load-balancing in a GPU particle in cell code

NASA Astrophysics Data System (ADS)

Rossi, Francesco; Londrillo, Pasquale; Sgattoni, Andrea; Sinigardi, Stefano; Turchetti, Giorgio

2012-12-01

We present `jasmine', an implementation of a fully relativistic, 3D, electromagnetic Particle-In-Cell (PIC) code, capable of running simulations in various laser plasma acceleration regimes on Graphics-Processing-Units (GPUs) HPC clusters. Standard energy/charge preserving FDTD-based algorithms have been implemented using double precision and quadratic (or arbitrary sized) shape functions for the particle weighting. When porting a PIC scheme to the GPU architecture (or, in general, a shared memory environment), the particle-to-grid operations (e.g. the evaluation of the current density) require special care to avoid memory inconsistencies and conflicts. Here we present a robust implementation of this operation that is efficient for any number of particles per cell and particle shape function order. Our algorithm exploits the exposed GPU memory hierarchy and avoids the use of atomic operations, which can hurt performance especially when many particles lay on the same cell. We show the code multi-GPU scalability results and present a dynamic load-balancing algorithm. The code is written using a python-based C++ meta-programming technique which translates in a high level of modularity and allows for easy performance tuning and simple extension of the core algorithms to various simulation schemes.

DOE Office of Scientific and Technical Information (OSTI.GOV)

Allada, Veerendra, Benjegerdes, Troy; Bode, Brett

Commodity clusters augmented with application accelerators are evolving as competitive high performance computing systems. The Graphical Processing Unit (GPU) with a very high arithmetic density and performance per price ratio is a good platform for the scientific application acceleration. In addition to the interconnect bottlenecks among the cluster compute nodes, the cost of memory copies between the host and the GPU device have to be carefully amortized to improve the overall efficiency of the application. Scientific applications also rely on efficient implementation of the BAsic Linear Algebra Subroutines (BLAS), among which the General Matrix Multiply (GEMM) is considered as themore » workhorse subroutine. In this paper, they study the performance of the memory copies and GEMM subroutines that are critical to port the computational chemistry algorithms to the GPU clusters. To that end, a benchmark based on the NetPIPE framework is developed to evaluate the latency and bandwidth of the memory copies between the host and the GPU device. The performance of the single and double precision GEMM subroutines from the NVIDIA CUBLAS 2.0 library are studied. The results have been compared with that of the BLAS routines from the Intel Math Kernel Library (MKL) to understand the computational trade-offs. The test bed is a Intel Xeon cluster equipped with NVIDIA Tesla GPUs.« less

Particle-in-cell simulations on graphic processing units

NASA Astrophysics Data System (ADS)

Ren, C.; Zhou, X.; Li, J.; Huang, M. C.; Zhao, Y.

2014-10-01

We will show our recent progress in using GPU's to accelerate the PIC code OSIRIS [Fonseca et al. LNCS 2331, 342 (2002)]. The OISRIS parallel structure is retained and the computation-intensive kernels are shipped to GPU's. Algorithms for the kernels are adapted for the GPU, including high-order charge-conserving current deposition schemes with few branching and parallel particle sorting [Kong et al., JCP 230, 1676 (2011)]. These algorithms make efficient use of the GPU shared memory. This work was supported by U.S. Department of Energy under Grant No. DE-FC02-04ER54789 and by NSF under Grant No. PHY-1314734.

GPU-accelerated phase-field simulation of dendritic solidification in a binary alloy

NASA Astrophysics Data System (ADS)

Yamanaka, Akinori; Aoki, Takayuki; Ogawa, Satoi; Takaki, Tomohiro

2011-03-01

The phase-field simulation for dendritic solidification of a binary alloy has been accelerated by using a graphic processing unit (GPU). To perform the phase-field simulation of the alloy solidification on GPU, a program code was developed with computer unified device architecture (CUDA). In this paper, the implementation technique of the phase-field model on GPU is presented. Also, we evaluated the acceleration performance of the three-dimensional solidification simulation by using a single NVIDIA TESLA C1060 GPU and the developed program code. The results showed that the GPU calculation for 5763 computational grids achieved the performance of 170 GFLOPS by utilizing the shared memory as a software-managed cache. Furthermore, it can be demonstrated that the computation with the GPU is 100 times faster than that with a single CPU core. From the obtained results, we confirmed the feasibility of realizing a real-time full three-dimensional phase-field simulation of microstructure evolution on a personal desktop computer.

GPU-accelerated algorithms for compressed signals recovery with application to astronomical imagery deblurring

NASA Astrophysics Data System (ADS)

Fiandrotti, Attilio; Fosson, Sophie M.; Ravazzi, Chiara; Magli, Enrico

2018-04-01

Compressive sensing promises to enable bandwidth-efficient on-board compression of astronomical data by lifting the encoding complexity from the source to the receiver. The signal is recovered off-line, exploiting GPUs parallel computation capabilities to speedup the reconstruction process. However, inherent GPU hardware constraints limit the size of the recoverable signal and the speedup practically achievable. In this work, we design parallel algorithms that exploit the properties of circulant matrices for efficient GPU-accelerated sparse signals recovery. Our approach reduces the memory requirements, allowing us to recover very large signals with limited memory. In addition, it achieves a tenfold signal recovery speedup thanks to ad-hoc parallelization of matrix-vector multiplications and matrix inversions. Finally, we practically demonstrate our algorithms in a typical application of circulant matrices: deblurring a sparse astronomical image in the compressed domain.

Practical Implementation of Prestack Kirchhoff Time Migration on a General Purpose Graphics Processing Unit

NASA Astrophysics Data System (ADS)

Liu, Guofeng; Li, Chun

2016-08-01

In this study, we present a practical implementation of prestack Kirchhoff time migration (PSTM) on a general purpose graphic processing unit. First, we consider the three main optimizations of the PSTM GPU code, i.e., designing a configuration based on a reasonable execution, using the texture memory for velocity interpolation, and the application of an intrinsic function in device code. This approach can achieve a speedup of nearly 45 times on a NVIDIA GTX 680 GPU compared with CPU code when a larger imaging space is used, where the PSTM output is a common reflection point that is gathered as I[ nx][ ny][ nh][ nt] in matrix format. However, this method requires more memory space so the limited imaging space cannot fully exploit the GPU sources. To overcome this problem, we designed a PSTM scheme with multi-GPUs for imaging different seismic data on different GPUs using an offset value. This process can achieve the peak speedup of GPU PSTM code and it greatly increases the efficiency of the calculations, but without changing the imaging result.

Viscoelastic Finite Difference Modeling Using Graphics Processing Units

NASA Astrophysics Data System (ADS)

Fabien-Ouellet, G.; Gloaguen, E.; Giroux, B.

2014-12-01

Full waveform seismic modeling requires a huge amount of computing power that still challenges today's technology. This limits the applicability of powerful processing approaches in seismic exploration like full-waveform inversion. This paper explores the use of Graphics Processing Units (GPU) to compute a time based finite-difference solution to the viscoelastic wave equation. The aim is to investigate whether the adoption of the GPU technology is susceptible to reduce significantly the computing time of simulations. The code presented herein is based on the freely accessible software of Bohlen (2002) in 2D provided under a General Public License (GNU) licence. This implementation is based on a second order centred differences scheme to approximate time differences and staggered grid schemes with centred difference of order 2, 4, 6, 8, and 12 for spatial derivatives. The code is fully parallel and is written using the Message Passing Interface (MPI), and it thus supports simulations of vast seismic models on a cluster of CPUs. To port the code from Bohlen (2002) on GPUs, the OpenCl framework was chosen for its ability to work on both CPUs and GPUs and its adoption by most of GPU manufacturers. In our implementation, OpenCL works in conjunction with MPI, which allows computations on a cluster of GPU for large-scale model simulations. We tested our code for model sizes between 1002 and 60002 elements. Comparison shows a decrease in computation time of more than two orders of magnitude between the GPU implementation run on a AMD Radeon HD 7950 and the CPU implementation run on a 2.26 GHz Intel Xeon Quad-Core. The speed-up varies depending on the order of the finite difference approximation and generally increases for higher orders. Increasing speed-ups are also obtained for increasing model size, which can be explained by kernel overheads and delays introduced by memory transfers to and from the GPU through the PCI-E bus. Those tests indicate that the GPU memory size and the slow memory transfers are the limiting factors of our GPU implementation. Those results show the benefits of using GPUs instead of CPUs for time based finite-difference seismic simulations. The reductions in computation time and in hardware costs are significant and open the door for new approaches in seismic inversion.

Bayer image parallel decoding based on GPU

NASA Astrophysics Data System (ADS)

Hu, Rihui; Xu, Zhiyong; Wei, Yuxing; Sun, Shaohua

2012-11-01

In the photoelectrical tracking system, Bayer image is decompressed in traditional method, which is CPU-based. However, it is too slow when the images become large, for example, 2K×2K×16bit. In order to accelerate the Bayer image decoding, this paper introduces a parallel speedup method for NVIDA's Graphics Processor Unit (GPU) which supports CUDA architecture. The decoding procedure can be divided into three parts: the first is serial part, the second is task-parallelism part, and the last is data-parallelism part including inverse quantization, inverse discrete wavelet transform (IDWT) as well as image post-processing part. For reducing the execution time, the task-parallelism part is optimized by OpenMP techniques. The data-parallelism part could advance its efficiency through executing on the GPU as CUDA parallel program. The optimization techniques include instruction optimization, shared memory access optimization, the access memory coalesced optimization and texture memory optimization. In particular, it can significantly speed up the IDWT by rewriting the 2D (Tow-dimensional) serial IDWT into 1D parallel IDWT. Through experimenting with 1K×1K×16bit Bayer image, data-parallelism part is 10 more times faster than CPU-based implementation. Finally, a CPU+GPU heterogeneous decompression system was designed. The experimental result shows that it could achieve 3 to 5 times speed increase compared to the CPU serial method.

DOE Office of Scientific and Technical Information (OSTI.GOV)

Lyakh, Dmitry I.

An efficient parallel tensor transpose algorithm is suggested for shared-memory computing units, namely, multicore CPU, Intel Xeon Phi, and NVidia GPU. The algorithm operates on dense tensors (multidimensional arrays) and is based on the optimization of cache utilization on x86 CPU and the use of shared memory on NVidia GPU. From the applied side, the ultimate goal is to minimize the overhead encountered in the transformation of tensor contractions into matrix multiplications in computer implementations of advanced methods of quantum many-body theory (e.g., in electronic structure theory and nuclear physics). A particular accent is made on higher-dimensional tensors that typicallymore » appear in the so-called multireference correlated methods of electronic structure theory. Depending on tensor dimensionality, the presented optimized algorithms can achieve an order of magnitude speedup on x86 CPUs and 2-3 times speedup on NVidia Tesla K20X GPU with respect to the na ve scattering algorithm (no memory access optimization). Furthermore, the tensor transpose routines developed in this work have been incorporated into a general-purpose tensor algebra library (TAL-SH).« less

Primal-dual convex optimization in large deformation diffeomorphic metric mapping: LDDMM meets robust regularizers

NASA Astrophysics Data System (ADS)

Hernandez, Monica

2017-12-01

This paper proposes a method for primal-dual convex optimization in variational large deformation diffeomorphic metric mapping problems formulated with robust regularizers and robust image similarity metrics. The method is based on Chambolle and Pock primal-dual algorithm for solving general convex optimization problems. Diagonal preconditioning is used to ensure the convergence of the algorithm to the global minimum. We consider three robust regularizers liable to provide acceptable results in diffeomorphic registration: Huber, V-Huber and total generalized variation. The Huber norm is used in the image similarity term. The primal-dual equations are derived for the stationary and the non-stationary parameterizations of diffeomorphisms. The resulting algorithms have been implemented for running in the GPU using Cuda. For the most memory consuming methods, we have developed a multi-GPU implementation. The GPU implementations allowed us to perform an exhaustive evaluation study in NIREP and LPBA40 databases. The experiments showed that, for all the considered regularizers, the proposed method converges to diffeomorphic solutions while better preserving discontinuities at the boundaries of the objects compared to baseline diffeomorphic registration methods. In most cases, the evaluation showed a competitive performance for the robust regularizers, close to the performance of the baseline diffeomorphic registration methods.

Benchmarking GPU and CPU codes for Heisenberg spin glass over-relaxation

NASA Astrophysics Data System (ADS)

Bernaschi, M.; Parisi, G.; Parisi, L.

2011-06-01

We present a set of possible implementations for Graphics Processing Units (GPU) of the Over-relaxation technique applied to the 3D Heisenberg spin glass model. The results show that a carefully tuned code can achieve more than 100 GFlops/s of sustained performance and update a single spin in about 0.6 nanoseconds. A multi-hit technique that exploits the GPU shared memory further reduces this time. Such results are compared with those obtained by means of a highly-tuned vector-parallel code on latest generation multi-core CPUs.

Simultaneous Range-Velocity Processing and SNR Analysis of AFIT’s Random Noise Radar

DTIC Science & Technology

2012-03-22

reducing the overall processing time. Two computers, equipped with NVIDIA ® GPUs, were used to process the col- 45 lected data. The specifications for each...gather the results back to the CPU. Another company , AccelerEyes®, has developed a product called Jacket® that claims to be better than the parallel...Number of Processing Cores 4 8 Processor Speed 3.33 GHz 3.07 GHz Installed Memory 48 GB 48 GB GPU Make NVIDIA NVIDIA GPU Model Tesla 1060 Tesla C2070 GPU

Graphics Processing Unit (GPU) Acceleration of the Goddard Earth Observing System Atmospheric Model

NASA Technical Reports Server (NTRS)

Putnam, Williama

2011-01-01

The Goddard Earth Observing System 5 (GEOS-5) is the atmospheric model used by the Global Modeling and Assimilation Office (GMAO) for a variety of applications, from long-term climate prediction at relatively coarse resolution, to data assimilation and numerical weather prediction, to very high-resolution cloud-resolving simulations. GEOS-5 is being ported to a graphics processing unit (GPU) cluster at the NASA Center for Climate Simulation (NCCS). By utilizing GPU co-processor technology, we expect to increase the throughput of GEOS-5 by at least an order of magnitude, and accelerate the process of scientific exploration across all scales of global modeling, including: The large-scale, high-end application of non-hydrostatic, global, cloud-resolving modeling at 10- to I-kilometer (km) global resolutions Intermediate-resolution seasonal climate and weather prediction at 50- to 25-km on small clusters of GPUs Long-range, coarse-resolution climate modeling, enabled on a small box of GPUs for the individual researcher After being ported to the GPU cluster, the primary physics components and the dynamical core of GEOS-5 have demonstrated a potential speedup of 15-40 times over conventional processor cores. Performance improvements of this magnitude reduce the required scalability of 1-km, global, cloud-resolving models from an unfathomable 6 million cores to an attainable 200,000 GPU-enabled cores.

GPU-based Parallel Application Design for Emerging Mobile Devices

NASA Astrophysics Data System (ADS)

Gupta, Kshitij

A revolution is underway in the computing world that is causing a fundamental paradigm shift in device capabilities and form-factor, with a move from well-established legacy desktop/laptop computers to mobile devices in varying sizes and shapes. Amongst all the tasks these devices must support, graphics has emerged as the 'killer app' for providing a fluid user interface and high-fidelity game rendering, effectively making the graphics processor (GPU) one of the key components in (present and future) mobile systems. By utilizing the GPU as a general-purpose parallel processor, this dissertation explores the GPU computing design space from an applications standpoint, in the mobile context, by focusing on key challenges presented by these devices---limited compute, memory bandwidth, and stringent power consumption requirements---while improving the overall application efficiency of the increasingly important speech recognition workload for mobile user interaction. We broadly partition trends in GPU computing into four major categories. We analyze hardware and programming model limitations in current-generation GPUs and detail an alternate programming style called Persistent Threads, identify four use case patterns, and propose minimal modifications that would be required for extending native support. We show how by manually extracting data locality and altering the speech recognition pipeline, we are able to achieve significant savings in memory bandwidth while simultaneously reducing the compute burden on GPU-like parallel processors. As we foresee GPU computing to evolve from its current 'co-processor' model into an independent 'applications processor' that is capable of executing complex work independently, we create an alternate application framework that enables the GPU to handle all control-flow dependencies autonomously at run-time while minimizing host involvement to just issuing commands, that facilitates an efficient application implementation. Finally, as compute and communication capabilities of mobile devices improve, we analyze energy implications of processing speech recognition locally (on-chip) and offloading it to servers (in-cloud).

GPU Accelerated Clustering for Arbitrary Shapes in Geoscience Data

NASA Astrophysics Data System (ADS)

Pankratius, V.; Gowanlock, M.; Rude, C. M.; Li, J. D.

2016-12-01

Clustering algorithms have become a vital component in intelligent systems for geoscience that helps scientists discover and track phenomena of various kinds. Here, we outline advances in Density-Based Spatial Clustering of Applications with Noise (DBSCAN) which detects clusters of arbitrary shape that are common in geospatial data. In particular, we propose a hybrid CPU-GPU implementation of DBSCAN and highlight new optimization approaches on the GPU that allows clustering detection in parallel while optimizing data transport during CPU-GPU interactions. We employ an efficient batching scheme between the host and GPU such that limited GPU memory is not prohibitive when processing large and/or dense datasets. To minimize data transfer overhead, we estimate the total workload size and employ an execution that generates optimized batches that will not overflow the GPU buffer. This work is demonstrated on space weather Total Electron Content (TEC) datasets containing over 5 million measurements from instruments worldwide, and allows scientists to spot spatially coherent phenomena with ease. Our approach is up to 30 times faster than a sequential implementation and therefore accelerates discoveries in large datasets. We acknowledge support from NSF ACI-1442997.

A real-time spike sorting method based on the embedded GPU.

PubMed

Zelan Yang; Kedi Xu; Xiang Tian; Shaomin Zhang; Xiaoxiang Zheng

2017-07-01

Microelectrode arrays with hundreds of channels have been widely used to acquire neuron population signals in neuroscience studies. Online spike sorting is becoming one of the most important challenges for high-throughput neural signal acquisition systems. Graphic processing unit (GPU) with high parallel computing capability might provide an alternative solution for increasing real-time computational demands on spike sorting. This study reported a method of real-time spike sorting through computing unified device architecture (CUDA) which was implemented on an embedded GPU (NVIDIA JETSON Tegra K1, TK1). The sorting approach is based on the principal component analysis (PCA) and K-means. By analyzing the parallelism of each process, the method was further optimized in the thread memory model of GPU. Our results showed that the GPU-based classifier on TK1 is 37.92 times faster than the MATLAB-based classifier on PC while their accuracies were the same with each other. The high-performance computing features of embedded GPU demonstrated in our studies suggested that the embedded GPU provide a promising platform for the real-time neural signal processing.

GPU accelerated manifold correction method for spinning compact binaries

NASA Astrophysics Data System (ADS)

Ran, Chong-xi; Liu, Song; Zhong, Shuang-ying

2018-04-01

The graphics processing unit (GPU) acceleration of the manifold correction algorithm based on the compute unified device architecture (CUDA) technology is designed to simulate the dynamic evolution of the Post-Newtonian (PN) Hamiltonian formulation of spinning compact binaries. The feasibility and the efficiency of parallel computation on GPU have been confirmed by various numerical experiments. The numerical comparisons show that the accuracy on GPU execution of manifold corrections method has a good agreement with the execution of codes on merely central processing unit (CPU-based) method. The acceleration ability when the codes are implemented on GPU can increase enormously through the use of shared memory and register optimization techniques without additional hardware costs, implying that the speedup is nearly 13 times as compared with the codes executed on CPU for phase space scan (including 314 × 314 orbits). In addition, GPU-accelerated manifold correction method is used to numerically study how dynamics are affected by the spin-induced quadrupole-monopole interaction for black hole binary system.

GPU COMPUTING FOR PARTICLE TRACKING

DOE Office of Scientific and Technical Information (OSTI.GOV)

Nishimura, Hiroshi; Song, Kai; Muriki, Krishna

2011-03-25

This is a feasibility study of using a modern Graphics Processing Unit (GPU) to parallelize the accelerator particle tracking code. To demonstrate the massive parallelization features provided by GPU computing, a simplified TracyGPU program is developed for dynamic aperture calculation. Performances, issues, and challenges from introducing GPU are also discussed. General purpose Computation on Graphics Processing Units (GPGPU) bring massive parallel computing capabilities to numerical calculation. However, the unique architecture of GPU requires a comprehensive understanding of the hardware and programming model to be able to well optimize existing applications. In the field of accelerator physics, the dynamic aperture calculationmore » of a storage ring, which is often the most time consuming part of the accelerator modeling and simulation, can benefit from GPU due to its embarrassingly parallel feature, which fits well with the GPU programming model. In this paper, we use the Tesla C2050 GPU which consists of 14 multi-processois (MP) with 32 cores on each MP, therefore a total of 448 cores, to host thousands ot threads dynamically. Thread is a logical execution unit of the program on GPU. In the GPU programming model, threads are grouped into a collection of blocks Within each block, multiple threads share the same code, and up to 48 KB of shared memory. Multiple thread blocks form a grid, which is executed as a GPU kernel. A simplified code that is a subset of Tracy++ [2] is developed to demonstrate the possibility of using GPU to speed up the dynamic aperture calculation by having each thread track a particle.« less

GPU-accelerated adjoint algorithmic differentiation

NASA Astrophysics Data System (ADS)

Gremse, Felix; Höfter, Andreas; Razik, Lukas; Kiessling, Fabian; Naumann, Uwe

2016-03-01

Many scientific problems such as classifier training or medical image reconstruction can be expressed as minimization of differentiable real-valued cost functions and solved with iterative gradient-based methods. Adjoint algorithmic differentiation (AAD) enables automated computation of gradients of such cost functions implemented as computer programs. To backpropagate adjoint derivatives, excessive memory is potentially required to store the intermediate partial derivatives on a dedicated data structure, referred to as the ;tape;. Parallelization is difficult because threads need to synchronize their accesses during taping and backpropagation. This situation is aggravated for many-core architectures, such as Graphics Processing Units (GPUs), because of the large number of light-weight threads and the limited memory size in general as well as per thread. We show how these limitations can be mediated if the cost function is expressed using GPU-accelerated vector and matrix operations which are recognized as intrinsic functions by our AAD software. We compare this approach with naive and vectorized implementations for CPUs. We use four increasingly complex cost functions to evaluate the performance with respect to memory consumption and gradient computation times. Using vectorization, CPU and GPU memory consumption could be substantially reduced compared to the naive reference implementation, in some cases even by an order of complexity. The vectorization allowed usage of optimized parallel libraries during forward and reverse passes which resulted in high speedups for the vectorized CPU version compared to the naive reference implementation. The GPU version achieved an additional speedup of 7.5 ± 4.4, showing that the processing power of GPUs can be utilized for AAD using this concept. Furthermore, we show how this software can be systematically extended for more complex problems such as nonlinear absorption reconstruction for fluorescence-mediated tomography.

GPU-Accelerated Adjoint Algorithmic Differentiation.

PubMed

Gremse, Felix; Höfter, Andreas; Razik, Lukas; Kiessling, Fabian; Naumann, Uwe

2016-03-01

Many scientific problems such as classifier training or medical image reconstruction can be expressed as minimization of differentiable real-valued cost functions and solved with iterative gradient-based methods. Adjoint algorithmic differentiation (AAD) enables automated computation of gradients of such cost functions implemented as computer programs. To backpropagate adjoint derivatives, excessive memory is potentially required to store the intermediate partial derivatives on a dedicated data structure, referred to as the "tape". Parallelization is difficult because threads need to synchronize their accesses during taping and backpropagation. This situation is aggravated for many-core architectures, such as Graphics Processing Units (GPUs), because of the large number of light-weight threads and the limited memory size in general as well as per thread. We show how these limitations can be mediated if the cost function is expressed using GPU-accelerated vector and matrix operations which are recognized as intrinsic functions by our AAD software. We compare this approach with naive and vectorized implementations for CPUs. We use four increasingly complex cost functions to evaluate the performance with respect to memory consumption and gradient computation times. Using vectorization, CPU and GPU memory consumption could be substantially reduced compared to the naive reference implementation, in some cases even by an order of complexity. The vectorization allowed usage of optimized parallel libraries during forward and reverse passes which resulted in high speedups for the vectorized CPU version compared to the naive reference implementation. The GPU version achieved an additional speedup of 7.5 ± 4.4, showing that the processing power of GPUs can be utilized for AAD using this concept. Furthermore, we show how this software can be systematically extended for more complex problems such as nonlinear absorption reconstruction for fluorescence-mediated tomography.

GPU-Accelerated Adjoint Algorithmic Differentiation

PubMed Central

Gremse, Felix; Höfter, Andreas; Razik, Lukas; Kiessling, Fabian; Naumann, Uwe

2015-01-01

Many scientific problems such as classifier training or medical image reconstruction can be expressed as minimization of differentiable real-valued cost functions and solved with iterative gradient-based methods. Adjoint algorithmic differentiation (AAD) enables automated computation of gradients of such cost functions implemented as computer programs. To backpropagate adjoint derivatives, excessive memory is potentially required to store the intermediate partial derivatives on a dedicated data structure, referred to as the “tape”. Parallelization is difficult because threads need to synchronize their accesses during taping and backpropagation. This situation is aggravated for many-core architectures, such as Graphics Processing Units (GPUs), because of the large number of light-weight threads and the limited memory size in general as well as per thread. We show how these limitations can be mediated if the cost function is expressed using GPU-accelerated vector and matrix operations which are recognized as intrinsic functions by our AAD software. We compare this approach with naive and vectorized implementations for CPUs. We use four increasingly complex cost functions to evaluate the performance with respect to memory consumption and gradient computation times. Using vectorization, CPU and GPU memory consumption could be substantially reduced compared to the naive reference implementation, in some cases even by an order of complexity. The vectorization allowed usage of optimized parallel libraries during forward and reverse passes which resulted in high speedups for the vectorized CPU version compared to the naive reference implementation. The GPU version achieved an additional speedup of 7.5 ± 4.4, showing that the processing power of GPUs can be utilized for AAD using this concept. Furthermore, we show how this software can be systematically extended for more complex problems such as nonlinear absorption reconstruction for fluorescence-mediated tomography. PMID:26941443

Storage strategies of eddy-current FE-BI model for GPU implementation

NASA Astrophysics Data System (ADS)

Bardel, Charles; Lei, Naiguang; Udpa, Lalita

2013-01-01

In the past few years graphical processing units (GPUs) have shown tremendous improvements in computational throughput over standard CPU architecture. However, this comes at the cost of restructuring the algorithms to meet the strengths and drawbacks of this GPU architecture. A major drawback is the state of limited memory, and hence storage of FE stiffness matrices on the GPU is important. In contrast to storage on CPU the GPU storage format has significant influence on the overall performance. This paper presents an investigation of a storage strategy in the implementation of a two-dimensional finite element-boundary integral (FE-BI) model for Eddy current NDE applications, on GPU architecture. Specifically, the high dimensional matrices are manipulated by examining the matrix structure and optimally splitting into structurally independent component matrices for efficient storage and retrieval of each component. Results obtained using the proposed approach are compared to those of conventional CPU implementation for validating the method.

GPU-Accelerated Stony-Brook University 5-class Microphysics Scheme in WRF

NASA Astrophysics Data System (ADS)

Mielikainen, J.; Huang, B.; Huang, A.

2011-12-01

The Weather Research and Forecasting (WRF) model is a next-generation mesoscale numerical weather prediction system. Microphysics plays an important role in weather and climate prediction. Several bulk water microphysics schemes are available within the WRF, with different numbers of simulated hydrometeor classes and methods for estimating their size fall speeds, distributions and densities. Stony-Brook University scheme (SBU-YLIN) is a 5-class scheme with riming intensity predicted to account for mixed-phase processes. In the past few years, co-processing on Graphics Processing Units (GPUs) has been a disruptive technology in High Performance Computing (HPC). GPUs use the ever increasing transistor count for adding more processor cores. Therefore, GPUs are well suited for massively data parallel processing with high floating point arithmetic intensity. Thus, it is imperative to update legacy scientific applications to take advantage of this unprecedented increase in computing power. CUDA is an extension to the C programming language offering programming GPU's directly. It is designed so that its constructs allow for natural expression of data-level parallelism. A CUDA program is organized into two parts: a serial program running on the CPU and a CUDA kernel running on the GPU. The CUDA code consists of three computational phases: transmission of data into the global memory of the GPU, execution of the CUDA kernel, and transmission of results from the GPU into the memory of CPU. CUDA takes a bottom-up point of view of parallelism is which thread is an atomic unit of parallelism. Individual threads are part of groups called warps, within which every thread executes exactly the same sequence of instructions. To test SBU-YLIN, we used a CONtinental United States (CONUS) benchmark data set for 12 km resolution domain for October 24, 2001. A WRF domain is a geographic region of interest discretized into a 2-dimensional grid parallel to the ground. Each grid point has multiple levels, which correspond to various vertical heights in the atmosphere. The size of the CONUS 12 km domain is 433 x 308 horizontal grid points with 35 vertical levels. First, the entire SBU-YLIN Fortran code was rewritten in C in preparation of GPU accelerated version. After that, C code was verified against Fortran code for identical outputs. Default compiler options from WRF were used for gfortran and gcc compilers. The processing time for the original Fortran code is 12274 ms and 12893 ms for C version. The processing times for GPU implementation of SBU-YLIN microphysics scheme with I/O are 57.7 ms and 37.2 ms for 1 and 2 GPUs, respectively. The corresponding speedups are 213x and 330x compared to a Fortran implementation. Without I/O the speedup is 896x on 1 GPU. Obviously, ignoring I/O time speedup scales linearly with GPUs. Thus, 2 GPUs have a speedup of 1788x without I/O. Microphysics computation is just a small part of the whole WRF model. After having completely implemented WRF on GPU, the inputs for SBU-YLIN do not have to be transferred from CPU. Instead they are results of previous WRF modules. Therefore, the role of I/O is greatly diminished once all of WRF have been converted to run on GPUs. In the near future, we expect to have a WRF running completely on GPUs for a superior performance.

GPU Implementation of High Rayleigh Number Three-Dimensional Mantle Convection

NASA Astrophysics Data System (ADS)

Sanchez, D. A.; Yuen, D. A.; Wright, G. B.; Barnett, G. A.

2010-12-01

Although we have entered the age of petascale computing, many factors are still prohibiting high-performance computing (HPC) from infiltrating all suitable scientific disciplines. For this reason and others, application of GPU to HPC is gaining traction in the scientific world. With its low price point, high performance potential, and competitive scalability, GPU has been an option well worth considering for the last few years. Moreover with the advent of NVIDIA's Fermi architecture, which brings ECC memory, better double-precision performance, and more RAM to GPU, there is a strong message of corporate support for GPU in HPC. However many doubts linger concerning the practicality of using GPU for scientific computing. In particular, GPU has a reputation for being difficult to program and suitable for only a small subset of problems. Although inroads have been made in addressing these concerns, for many scientists GPU still has hurdles to clear before becoming an acceptable choice. We explore the applicability of GPU to geophysics by implementing a three-dimensional, second-order finite-difference model of Rayleigh-Benard thermal convection on an NVIDIA GPU using C for CUDA. Our code reaches sufficient resolution, on the order of 500x500x250 evenly-spaced finite-difference gridpoints, on a single GPU. We make extensive use of highly optimized CUBLAS routines, allowing us to achieve performance on the order of O( 0.1 ) µs per timestep*gridpoint at this resolution. This performance has allowed us to study high Rayleigh number simulations, on the order of 2x10^7, on a single GPU.

Hardware Acceleration of Sparse Cognitive Algorithms

DTIC Science & Technology

2016-05-01

Processor in Memory (PiM) extensions and a 65 nm ASIC version. They were compared against a 28 nm GPU baseline using the KTH video action recognition...30 Table 17. Memory Requirement of Proposed ASIC...for improvement of performance per unit of power for customized implementations of the Sparsey and Numenta Hierarchical Temporal Memory (HTM

Accelerated rescaling of single Monte Carlo simulation runs with the Graphics Processing Unit (GPU).

PubMed

Yang, Owen; Choi, Bernard

2013-01-01

To interpret fiber-based and camera-based measurements of remitted light from biological tissues, researchers typically use analytical models, such as the diffusion approximation to light transport theory, or stochastic models, such as Monte Carlo modeling. To achieve rapid (ideally real-time) measurement of tissue optical properties, especially in clinical situations, there is a critical need to accelerate Monte Carlo simulation runs. In this manuscript, we report on our approach using the Graphics Processing Unit (GPU) to accelerate rescaling of single Monte Carlo runs to calculate rapidly diffuse reflectance values for different sets of tissue optical properties. We selected MATLAB to enable non-specialists in C and CUDA-based programming to use the generated open-source code. We developed a software package with four abstraction layers. To calculate a set of diffuse reflectance values from a simulated tissue with homogeneous optical properties, our rescaling GPU-based approach achieves a reduction in computation time of several orders of magnitude as compared to other GPU-based approaches. Specifically, our GPU-based approach generated a diffuse reflectance value in 0.08ms. The transfer time from CPU to GPU memory currently is a limiting factor with GPU-based calculations. However, for calculation of multiple diffuse reflectance values, our GPU-based approach still can lead to processing that is ~3400 times faster than other GPU-based approaches.

JiTTree: A Just-in-Time Compiled Sparse GPU Volume Data Structure.

PubMed

Labschütz, Matthias; Bruckner, Stefan; Gröller, M Eduard; Hadwiger, Markus; Rautek, Peter

2016-01-01

Sparse volume data structures enable the efficient representation of large but sparse volumes in GPU memory for computation and visualization. However, the choice of a specific data structure for a given data set depends on several factors, such as the memory budget, the sparsity of the data, and data access patterns. In general, there is no single optimal sparse data structure, but a set of several candidates with individual strengths and drawbacks. One solution to this problem are hybrid data structures which locally adapt themselves to the sparsity. However, they typically suffer from increased traversal overhead which limits their utility in many applications. This paper presents JiTTree, a novel sparse hybrid volume data structure that uses just-in-time compilation to overcome these problems. By combining multiple sparse data structures and reducing traversal overhead we leverage their individual advantages. We demonstrate that hybrid data structures adapt well to a large range of data sets. They are especially superior to other sparse data structures for data sets that locally vary in sparsity. Possible optimization criteria are memory, performance and a combination thereof. Through just-in-time (JIT) compilation, JiTTree reduces the traversal overhead of the resulting optimal data structure. As a result, our hybrid volume data structure enables efficient computations on the GPU, while being superior in terms of memory usage when compared to non-hybrid data structures.

SU-F-T-256: 4D IMRT Planning Using An Early Prototype GPU-Enabled Eclipse Workstation

DOE Office of Scientific and Technical Information (OSTI.GOV)

Hagan, A; Modiri, A; Sawant, A

Purpose: True 4D IMRT planning, based on simultaneous spatiotemporal optimization has been shown to significantly improve plan quality in lung radiotherapy. However, the high computational complexity associated with such planning represents a significant barrier to widespread clinical deployment. We introduce an early prototype GPU-enabled Eclipse workstation for inverse planning. To our knowledge, this is the first GPUintegrated Eclipse system demonstrating the potential for clinical translation of GPU computing on a major commercially-available TPS. Methods: The prototype system comprised of four NVIDIA Tesla K80 GPUs, with a maximum processing capability of 8.5 Tflops per K80 card. The system architecture consisted ofmore » three key modules: (i) a GPU-based inverse planning module using a highly-parallelizable, swarm intelligence-based global optimization algorithm, (ii) a GPU-based open-source b-spline deformable image registration module, Elastix, and (iii) a CUDA-based data management module. For evaluation, aperture fluence weights in an IMRT plan were optimized over 9 beams,166 apertures and 10 respiratory phases (14940 variables) for a lung cancer case (GTV = 95 cc, right lower lobe, 15 mm cranio-caudal motion). Sensitivity of the planning time and memory expense to parameter variations was quantified. Results: GPU-based inverse planning was significantly accelerated compared to its CPU counterpart (36 vs 488 min, for 10 phases, 10 search agents and 10 iterations). The optimized IMRT plan significantly improved OAR sparing compared to the original internal target volume (ITV)-based clinical plan, while maintaining prescribed tumor coverage. The dose-sparing improvements were: Esophagus Dmax 50%, Heart Dmax 42% and Spinal cord Dmax 25%. Conclusion: Our early prototype system demonstrates that through massive parallelization, computationally intense tasks such as 4D treatment planning can be accomplished in clinically feasible timeframes. With further optimization, such systems are expected to enable the eventual clinical translation of higher-dimensional and complex treatment planning strategies to significantly improve plan quality. This work was partially supported through research funding from National Institutes of Health (R01CA169102) and Varian Medical Systems, Palo Alto, CA, USA.« less

GPU-accelerated computing for Lagrangian coherent structures of multi-body gravitational regimes

NASA Astrophysics Data System (ADS)

Lin, Mingpei; Xu, Ming; Fu, Xiaoyu

2017-04-01

Based on a well-established theoretical foundation, Lagrangian Coherent Structures (LCSs) have elicited widespread research on the intrinsic structures of dynamical systems in many fields, including the field of astrodynamics. Although the application of LCSs in dynamical problems seems straightforward theoretically, its associated computational cost is prohibitive. We propose a block decomposition algorithm developed on Compute Unified Device Architecture (CUDA) platform for the computation of the LCSs of multi-body gravitational regimes. In order to take advantage of GPU's outstanding computing properties, such as Shared Memory, Constant Memory, and Zero-Copy, the algorithm utilizes a block decomposition strategy to facilitate computation of finite-time Lyapunov exponent (FTLE) fields of arbitrary size and timespan. Simulation results demonstrate that this GPU-based algorithm can satisfy double-precision accuracy requirements and greatly decrease the time needed to calculate final results, increasing speed by approximately 13 times. Additionally, this algorithm can be generalized to various large-scale computing problems, such as particle filters, constellation design, and Monte-Carlo simulation.

The Development of WARP - A Framework for Continuous Energy Monte Carlo Neutron Transport in General 3D Geometries on GPUs

NASA Astrophysics Data System (ADS)

Bergmann, Ryan

Graphics processing units, or GPUs, have gradually increased in computational power from the small, job-specific boards of the early 1990s to the programmable powerhouses of today. Compared to more common central processing units, or CPUs, GPUs have a higher aggregate memory bandwidth, much higher floating-point operations per second (FLOPS), and lower energy consumption per FLOP. Because one of the main obstacles in exascale computing is power consumption, many new supercomputing platforms are gaining much of their computational capacity by incorporating GPUs into their compute nodes. Since CPU-optimized parallel algorithms are not directly portable to GPU architectures (or at least not without losing substantial performance), transport codes need to be rewritten to execute efficiently on GPUs. Unless this is done, reactor simulations cannot take full advantage of these new supercomputers. WARP, which can stand for ``Weaving All the Random Particles,'' is a three-dimensional (3D) continuous energy Monte Carlo neutron transport code developed in this work as to efficiently implement a continuous energy Monte Carlo neutron transport algorithm on a GPU. WARP accelerates Monte Carlo simulations while preserving the benefits of using the Monte Carlo Method, namely, very few physical and geometrical simplifications. WARP is able to calculate multiplication factors, flux tallies, and fission source distributions for time-independent problems, and can run in both criticality or fixed source modes. WARP can transport neutrons in unrestricted arrangements of parallelepipeds, hexagonal prisms, cylinders, and spheres. WARP uses an event-based algorithm, but with some important differences. Moving data is expensive, so WARP uses a remapping vector of pointer/index pairs to direct GPU threads to the data they need to access. The remapping vector is sorted by reaction type after every transport iteration using a high-efficiency parallel radix sort, which serves to keep the reaction types as contiguous as possible and removes completed histories from the transport cycle. The sort reduces the amount of divergence in GPU ``thread blocks,'' keeps the SIMD units as full as possible, and eliminates using memory bandwidth to check if a neutron in the batch has been terminated or not. Using a remapping vector means the data access pattern is irregular, but this is mitigated by using large batch sizes where the GPU can effectively eliminate the high cost of irregular global memory access. WARP modifies the standard unionized energy grid implementation to reduce memory traffic. Instead of storing a matrix of pointers indexed by reaction type and energy, WARP stores three matrices. The first contains cross section values, the second contains pointers to angular distributions, and a third contains pointers to energy distributions. This linked list type of layout increases memory usage, but lowers the number of data loads that are needed to determine a reaction by eliminating a pointer load to find a cross section value. Optimized, high-performance GPU code libraries are also used by WARP wherever possible. The CUDA performance primitives (CUDPP) library is used to perform the parallel reductions, sorts and sums, the CURAND library is used to seed the linear congruential random number generators, and the OptiX ray tracing framework is used for geometry representation. OptiX is a highly-optimized library developed by NVIDIA that automatically builds hierarchical acceleration structures around user-input geometry so only surfaces along a ray line need to be queried in ray tracing. WARP also performs material and cell number queries with OptiX by using a point-in-polygon like algorithm. WARP has shown that GPUs are an effective platform for performing Monte Carlo neutron transport with continuous energy cross sections. Currently, WARP is the most detailed and feature-rich program in existence for performing continuous energy Monte Carlo neutron transport in general 3D geometries on GPUs, but compared to production codes like Serpent and MCNP, WARP has limited capabilities. Despite WARP's lack of features, its novel algorithm implementations show that high performance can be achieved on a GPU despite the inherently divergent program flow and sparse data access patterns. WARP is not ready for everyday nuclear reactor calculations, but is a good platform for further development of GPU-accelerated Monte Carlo neutron transport. In it's current state, it may be a useful tool for multiplication factor searches, i.e. determining reactivity coefficients by perturbing material densities or temperatures, since these types of calculations typically do not require many flux tallies. (Abstract shortened by UMI.)

A framework for interactive visual analysis of heterogeneous marine data in an integrated problem solving environment

NASA Astrophysics Data System (ADS)

Liu, Shuai; Chen, Ge; Yao, Shifeng; Tian, Fenglin; Liu, Wei

2017-07-01

This paper presents a novel integrated marine visualization framework which focuses on processing, analyzing the multi-dimension spatiotemporal marine data in one workflow. Effective marine data visualization is needed in terms of extracting useful patterns, recognizing changes, and understanding physical processes in oceanography researches. However, the multi-source, multi-format, multi-dimension characteristics of marine data pose a challenge for interactive and feasible (timely) marine data analysis and visualization in one workflow. And, global multi-resolution virtual terrain environment is also needed to give oceanographers and the public a real geographic background reference and to help them to identify the geographical variation of ocean phenomena. This paper introduces a data integration and processing method to efficiently visualize and analyze the heterogeneous marine data. Based on the data we processed, several GPU-based visualization methods are explored to interactively demonstrate marine data. GPU-tessellated global terrain rendering using ETOPO1 data is realized and the video memory usage is controlled to ensure high efficiency. A modified ray-casting algorithm for the uneven multi-section Argo volume data is also presented and the transfer function is designed to analyze the 3D structure of ocean phenomena. Based on the framework we designed, an integrated visualization system is realized. The effectiveness and efficiency of the framework is demonstrated. This system is expected to make a significant contribution to the demonstration and understanding of marine physical process in a virtual global environment.

GPU-accelerated atmospheric chemical kinetics in the ECHAM/MESSy (EMAC) Earth system model (version 2.52)

NASA Astrophysics Data System (ADS)

Alvanos, Michail; Christoudias, Theodoros

2017-10-01

This paper presents an application of GPU accelerators in Earth system modeling. We focus on atmospheric chemical kinetics, one of the most computationally intensive tasks in climate-chemistry model simulations. We developed a software package that automatically generates CUDA kernels to numerically integrate atmospheric chemical kinetics in the global climate model ECHAM/MESSy Atmospheric Chemistry (EMAC), used to study climate change and air quality scenarios. A source-to-source compiler outputs a CUDA-compatible kernel by parsing the FORTRAN code generated by the Kinetic PreProcessor (KPP) general analysis tool. All Rosenbrock methods that are available in the KPP numerical library are supported.Performance evaluation, using Fermi and Pascal CUDA-enabled GPU accelerators, shows achieved speed-ups of 4. 5 × and 20. 4 × , respectively, of the kernel execution time. A node-to-node real-world production performance comparison shows a 1. 75 × speed-up over the non-accelerated application using the KPP three-stage Rosenbrock solver. We provide a detailed description of the code optimizations used to improve the performance including memory optimizations, control code simplification, and reduction of idle time. The accuracy and correctness of the accelerated implementation are evaluated by comparing to the CPU-only code of the application. The median relative difference is found to be less than 0.000000001 % when comparing the output of the accelerated kernel the CPU-only code.The approach followed, including the computational workload division, and the developed GPU solver code can potentially be used as the basis for hardware acceleration of numerous geoscientific models that rely on KPP for atmospheric chemical kinetics applications.

Fast quantum Monte Carlo on a GPU

NASA Astrophysics Data System (ADS)

Lutsyshyn, Y.

2015-02-01

We present a scheme for the parallelization of quantum Monte Carlo method on graphical processing units, focusing on variational Monte Carlo simulation of bosonic systems. We use asynchronous execution schemes with shared memory persistence, and obtain an excellent utilization of the accelerator. The CUDA code is provided along with a package that simulates liquid helium-4. The program was benchmarked on several models of Nvidia GPU, including Fermi GTX560 and M2090, and the Kepler architecture K20 GPU. Special optimization was developed for the Kepler cards, including placement of data structures in the register space of the Kepler GPUs. Kepler-specific optimization is discussed.

Graphics Processing Units for HEP trigger systems

NASA Astrophysics Data System (ADS)

Ammendola, R.; Bauce, M.; Biagioni, A.; Chiozzi, S.; Cotta Ramusino, A.; Fantechi, R.; Fiorini, M.; Giagu, S.; Gianoli, A.; Lamanna, G.; Lonardo, A.; Messina, A.; Neri, I.; Paolucci, P. S.; Piandani, R.; Pontisso, L.; Rescigno, M.; Simula, F.; Sozzi, M.; Vicini, P.

2016-07-01

General-purpose computing on GPUs (Graphics Processing Units) is emerging as a new paradigm in several fields of science, although so far applications have been tailored to the specific strengths of such devices as accelerator in offline computation. With the steady reduction of GPU latencies, and the increase in link and memory throughput, the use of such devices for real-time applications in high-energy physics data acquisition and trigger systems is becoming ripe. We will discuss the use of online parallel computing on GPU for synchronous low level trigger, focusing on CERN NA62 experiment trigger system. The use of GPU in higher level trigger system is also briefly considered.

Accelerating Smith-Waterman Algorithm for Biological Database Search on CUDA-Compatible GPUs

NASA Astrophysics Data System (ADS)

Munekawa, Yuma; Ino, Fumihiko; Hagihara, Kenichi

This paper presents a fast method capable of accelerating the Smith-Waterman algorithm for biological database search on a cluster of graphics processing units (GPUs). Our method is implemented using compute unified device architecture (CUDA), which is available on the nVIDIA GPU. As compared with previous methods, our method has four major contributions. (1) The method efficiently uses on-chip shared memory to reduce the data amount being transferred between off-chip video memory and processing elements in the GPU. (2) It also reduces the number of data fetches by applying a data reuse technique to query and database sequences. (3) A pipelined method is also implemented to overlap GPU execution with database access. (4) Finally, a master/worker paradigm is employed to accelerate hundreds of database searches on a cluster system. In experiments, the peak performance on a GeForce GTX 280 card reaches 8.32 giga cell updates per second (GCUPS). We also find that our method reduces the amount of data fetches to 1/140, achieving approximately three times higher performance than a previous CUDA-based method. Our 32-node cluster version is approximately 28 times faster than a single GPU version. Furthermore, the effective performance reaches 75.6 giga instructions per second (GIPS) using 32 GeForce 8800 GTX cards.

Multi-GPU implementation of a VMAT treatment plan optimization algorithm.

PubMed

Tian, Zhen; Peng, Fei; Folkerts, Michael; Tan, Jun; Jia, Xun; Jiang, Steve B

2015-06-01

Volumetric modulated arc therapy (VMAT) optimization is a computationally challenging problem due to its large data size, high degrees of freedom, and many hardware constraints. High-performance graphics processing units (GPUs) have been used to speed up the computations. However, GPU's relatively small memory size cannot handle cases with a large dose-deposition coefficient (DDC) matrix in cases of, e.g., those with a large target size, multiple targets, multiple arcs, and/or small beamlet size. The main purpose of this paper is to report an implementation of a column-generation-based VMAT algorithm, previously developed in the authors' group, on a multi-GPU platform to solve the memory limitation problem. While the column-generation-based VMAT algorithm has been previously developed, the GPU implementation details have not been reported. Hence, another purpose is to present detailed techniques employed for GPU implementation. The authors also would like to utilize this particular problem as an example problem to study the feasibility of using a multi-GPU platform to solve large-scale problems in medical physics. The column-generation approach generates VMAT apertures sequentially by solving a pricing problem (PP) and a master problem (MP) iteratively. In the authors' method, the sparse DDC matrix is first stored on a CPU in coordinate list format (COO). On the GPU side, this matrix is split into four submatrices according to beam angles, which are stored on four GPUs in compressed sparse row format. Computation of beamlet price, the first step in PP, is accomplished using multi-GPUs. A fast inter-GPU data transfer scheme is accomplished using peer-to-peer access. The remaining steps of PP and MP problems are implemented on CPU or a single GPU due to their modest problem scale and computational loads. Barzilai and Borwein algorithm with a subspace step scheme is adopted here to solve the MP problem. A head and neck (H&N) cancer case is then used to validate the authors' method. The authors also compare their multi-GPU implementation with three different single GPU implementation strategies, i.e., truncating DDC matrix (S1), repeatedly transferring DDC matrix between CPU and GPU (S2), and porting computations involving DDC matrix to CPU (S3), in terms of both plan quality and computational efficiency. Two more H&N patient cases and three prostate cases are used to demonstrate the advantages of the authors' method. The authors' multi-GPU implementation can finish the optimization process within ∼ 1 min for the H&N patient case. S1 leads to an inferior plan quality although its total time was 10 s shorter than the multi-GPU implementation due to the reduced matrix size. S2 and S3 yield the same plan quality as the multi-GPU implementation but take ∼4 and ∼6 min, respectively. High computational efficiency was consistently achieved for the other five patient cases tested, with VMAT plans of clinically acceptable quality obtained within 23-46 s. Conversely, to obtain clinically comparable or acceptable plans for all six of these VMAT cases that the authors have tested in this paper, the optimization time needed in a commercial TPS system on CPU was found to be in an order of several minutes. The results demonstrate that the multi-GPU implementation of the authors' column-generation-based VMAT optimization can handle the large-scale VMAT optimization problem efficiently without sacrificing plan quality. The authors' study may serve as an example to shed some light on other large-scale medical physics problems that require multi-GPU techniques.

Fast analysis of molecular dynamics trajectories with graphics processing units-Radial distribution function histogramming

DOE Office of Scientific and Technical Information (OSTI.GOV)

Levine, Benjamin G., E-mail: ben.levine@temple.ed; Stone, John E., E-mail: johns@ks.uiuc.ed; Kohlmeyer, Axel, E-mail: akohlmey@temple.ed

2011-05-01

The calculation of radial distribution functions (RDFs) from molecular dynamics trajectory data is a common and computationally expensive analysis task. The rate limiting step in the calculation of the RDF is building a histogram of the distance between atom pairs in each trajectory frame. Here we present an implementation of this histogramming scheme for multiple graphics processing units (GPUs). The algorithm features a tiling scheme to maximize the reuse of data at the fastest levels of the GPU's memory hierarchy and dynamic load balancing to allow high performance on heterogeneous configurations of GPUs. Several versions of the RDF algorithm aremore » presented, utilizing the specific hardware features found on different generations of GPUs. We take advantage of larger shared memory and atomic memory operations available on state-of-the-art GPUs to accelerate the code significantly. The use of atomic memory operations allows the fast, limited-capacity on-chip memory to be used much more efficiently, resulting in a fivefold increase in performance compared to the version of the algorithm without atomic operations. The ultimate version of the algorithm running in parallel on four NVIDIA GeForce GTX 480 (Fermi) GPUs was found to be 92 times faster than a multithreaded implementation running on an Intel Xeon 5550 CPU. On this multi-GPU hardware, the RDF between two selections of 1,000,000 atoms each can be calculated in 26.9 s per frame. The multi-GPU RDF algorithms described here are implemented in VMD, a widely used and freely available software package for molecular dynamics visualization and analysis.« less

Accelerating image reconstruction in dual-head PET system by GPU and symmetry properties.

PubMed

Chou, Cheng-Ying; Dong, Yun; Hung, Yukai; Kao, Yu-Jiun; Wang, Weichung; Kao, Chien-Min; Chen, Chin-Tu

2012-01-01

Positron emission tomography (PET) is an important imaging modality in both clinical usage and research studies. We have developed a compact high-sensitivity PET system that consisted of two large-area panel PET detector heads, which produce more than 224 million lines of response and thus request dramatic computational demands. In this work, we employed a state-of-the-art graphics processing unit (GPU), NVIDIA Tesla C2070, to yield an efficient reconstruction process. Our approaches ingeniously integrate the distinguished features of the symmetry properties of the imaging system and GPU architectures, including block/warp/thread assignments and effective memory usage, to accelerate the computations for ordered subset expectation maximization (OSEM) image reconstruction. The OSEM reconstruction algorithms were implemented employing both CPU-based and GPU-based codes, and their computational performance was quantitatively analyzed and compared. The results showed that the GPU-accelerated scheme can drastically reduce the reconstruction time and thus can largely expand the applicability of the dual-head PET system.

GPU computing with Kaczmarz’s and other iterative algorithms for linear systems

PubMed Central

Elble, Joseph M.; Sahinidis, Nikolaos V.; Vouzis, Panagiotis

2009-01-01

The graphics processing unit (GPU) is used to solve large linear systems derived from partial differential equations. The differential equations studied are strongly convection-dominated, of various sizes, and common to many fields, including computational fluid dynamics, heat transfer, and structural mechanics. The paper presents comparisons between GPU and CPU implementations of several well-known iterative methods, including Kaczmarz’s, Cimmino’s, component averaging, conjugate gradient normal residual (CGNR), symmetric successive overrelaxation-preconditioned conjugate gradient, and conjugate-gradient-accelerated component-averaged row projections (CARP-CG). Computations are preformed with dense as well as general banded systems. The results demonstrate that our GPU implementation outperforms CPU implementations of these algorithms, as well as previously studied parallel implementations on Linux clusters and shared memory systems. While the CGNR method had begun to fall out of favor for solving such problems, for the problems studied in this paper, the CGNR method implemented on the GPU performed better than the other methods, including a cluster implementation of the CARP-CG method. PMID:20526446

Protein-protein docking on hardware accelerators: comparison of GPU and MIC architectures

PubMed Central

2015-01-01

Background The hardware accelerators will provide solutions to computationally complex problems in bioinformatics fields. However, the effect of acceleration depends on the nature of the application, thus selection of an appropriate accelerator requires some consideration. Results In the present study, we compared the effects of acceleration using graphics processing unit (GPU) and many integrated core (MIC) on the speed of fast Fourier transform (FFT)-based protein-protein docking calculation. The GPU implementation performed the protein-protein docking calculations approximately five times faster than the MIC offload mode implementation. The MIC native mode implementation has the advantage in the implementation costs. However, the performance was worse with larger protein pairs because of memory limitations. Conclusion The results suggest that GPU is more suitable than MIC for accelerating FFT-based protein-protein docking applications. PMID:25707855

Parallel Implementation of MAFFT on CUDA-Enabled Graphics Hardware.

PubMed

Zhu, Xiangyuan; Li, Kenli; Salah, Ahmad; Shi, Lin; Li, Keqin

2015-01-01

Multiple sequence alignment (MSA) constitutes an extremely powerful tool for many biological applications including phylogenetic tree estimation, secondary structure prediction, and critical residue identification. However, aligning large biological sequences with popular tools such as MAFFT requires long runtimes on sequential architectures. Due to the ever increasing sizes of sequence databases, there is increasing demand to accelerate this task. In this paper, we demonstrate how graphic processing units (GPUs), powered by the compute unified device architecture (CUDA), can be used as an efficient computational platform to accelerate the MAFFT algorithm. To fully exploit the GPU's capabilities for accelerating MAFFT, we have optimized the sequence data organization to eliminate the bandwidth bottleneck of memory access, designed a memory allocation and reuse strategy to make full use of limited memory of GPUs, proposed a new modified-run-length encoding (MRLE) scheme to reduce memory consumption, and used high-performance shared memory to speed up I/O operations. Our implementation tested in three NVIDIA GPUs achieves speedup up to 11.28 on a Tesla K20m GPU compared to the sequential MAFFT 7.015.

Accelerating Spaceborne SAR Imaging Using Multiple CPU/GPU Deep Collaborative Computing

PubMed Central

Zhang, Fan; Li, Guojun; Li, Wei; Hu, Wei; Hu, Yuxin

2016-01-01

With the development of synthetic aperture radar (SAR) technologies in recent years, the huge amount of remote sensing data brings challenges for real-time imaging processing. Therefore, high performance computing (HPC) methods have been presented to accelerate SAR imaging, especially the GPU based methods. In the classical GPU based imaging algorithm, GPU is employed to accelerate image processing by massive parallel computing, and CPU is only used to perform the auxiliary work such as data input/output (IO). However, the computing capability of CPU is ignored and underestimated. In this work, a new deep collaborative SAR imaging method based on multiple CPU/GPU is proposed to achieve real-time SAR imaging. Through the proposed tasks partitioning and scheduling strategy, the whole image can be generated with deep collaborative multiple CPU/GPU computing. In the part of CPU parallel imaging, the advanced vector extension (AVX) method is firstly introduced into the multi-core CPU parallel method for higher efficiency. As for the GPU parallel imaging, not only the bottlenecks of memory limitation and frequent data transferring are broken, but also kinds of optimized strategies are applied, such as streaming, parallel pipeline and so on. Experimental results demonstrate that the deep CPU/GPU collaborative imaging method enhances the efficiency of SAR imaging on single-core CPU by 270 times and realizes the real-time imaging in that the imaging rate outperforms the raw data generation rate. PMID:27070606

Accelerating Spaceborne SAR Imaging Using Multiple CPU/GPU Deep Collaborative Computing.

PubMed

Zhang, Fan; Li, Guojun; Li, Wei; Hu, Wei; Hu, Yuxin

2016-04-07

With the development of synthetic aperture radar (SAR) technologies in recent years, the huge amount of remote sensing data brings challenges for real-time imaging processing. Therefore, high performance computing (HPC) methods have been presented to accelerate SAR imaging, especially the GPU based methods. In the classical GPU based imaging algorithm, GPU is employed to accelerate image processing by massive parallel computing, and CPU is only used to perform the auxiliary work such as data input/output (IO). However, the computing capability of CPU is ignored and underestimated. In this work, a new deep collaborative SAR imaging method based on multiple CPU/GPU is proposed to achieve real-time SAR imaging. Through the proposed tasks partitioning and scheduling strategy, the whole image can be generated with deep collaborative multiple CPU/GPU computing. In the part of CPU parallel imaging, the advanced vector extension (AVX) method is firstly introduced into the multi-core CPU parallel method for higher efficiency. As for the GPU parallel imaging, not only the bottlenecks of memory limitation and frequent data transferring are broken, but also kinds of optimized strategies are applied, such as streaming, parallel pipeline and so on. Experimental results demonstrate that the deep CPU/GPU collaborative imaging method enhances the efficiency of SAR imaging on single-core CPU by 270 times and realizes the real-time imaging in that the imaging rate outperforms the raw data generation rate.

GASPRNG: GPU accelerated scalable parallel random number generator library

NASA Astrophysics Data System (ADS)

Gao, Shuang; Peterson, Gregory D.

2013-04-01

Graphics processors represent a promising technology for accelerating computational science applications. Many computational science applications require fast and scalable random number generation with good statistical properties, so they use the Scalable Parallel Random Number Generators library (SPRNG). We present the GPU Accelerated SPRNG library (GASPRNG) to accelerate SPRNG in GPU-based high performance computing systems. GASPRNG includes code for a host CPU and CUDA code for execution on NVIDIA graphics processing units (GPUs) along with a programming interface to support various usage models for pseudorandom numbers and computational science applications executing on the CPU, GPU, or both. This paper describes the implementation approach used to produce high performance and also describes how to use the programming interface. The programming interface allows a user to be able to use GASPRNG the same way as SPRNG on traditional serial or parallel computers as well as to develop tightly coupled programs executing primarily on the GPU. We also describe how to install GASPRNG and use it. To help illustrate linking with GASPRNG, various demonstration codes are included for the different usage models. GASPRNG on a single GPU shows up to 280x speedup over SPRNG on a single CPU core and is able to scale for larger systems in the same manner as SPRNG. Because GASPRNG generates identical streams of pseudorandom numbers as SPRNG, users can be confident about the quality of GASPRNG for scalable computational science applications. Catalogue identifier: AEOI_v1_0 Program summary URL:http://cpc.cs.qub.ac.uk/summaries/AEOI_v1_0.html Program obtainable from: CPC Program Library, Queen’s University, Belfast, N. Ireland Licensing provisions: UTK license. No. of lines in distributed program, including test data, etc.: 167900 No. of bytes in distributed program, including test data, etc.: 1422058 Distribution format: tar.gz Programming language: C and CUDA. Computer: Any PC or workstation with NVIDIA GPU (Tested on Fermi GTX480, Tesla C1060, Tesla M2070). Operating system: Linux with CUDA version 4.0 or later. Should also run on MacOS, Windows, or UNIX. Has the code been vectorized or parallelized?: Yes. Parallelized using MPI directives. RAM: 512 MB˜ 732 MB (main memory on host CPU, depending on the data type of random numbers.) / 512 MB (GPU global memory) Classification: 4.13, 6.5. Nature of problem: Many computational science applications are able to consume large numbers of random numbers. For example, Monte Carlo simulations are able to consume limitless random numbers for the computation as long as resources for the computing are supported. Moreover, parallel computational science applications require independent streams of random numbers to attain statistically significant results. The SPRNG library provides this capability, but at a significant computational cost. The GASPRNG library presented here accelerates the generators of independent streams of random numbers using graphical processing units (GPUs). Solution method: Multiple copies of random number generators in GPUs allow a computational science application to consume large numbers of random numbers from independent, parallel streams. GASPRNG is a random number generators library to allow a computational science application to employ multiple copies of random number generators to boost performance. Users can interface GASPRNG with software code executing on microprocessors and/or GPUs. Running time: The tests provided take a few minutes to run.

High performance computing for deformable image registration: towards a new paradigm in adaptive radiotherapy.

PubMed

Samant, Sanjiv S; Xia, Junyi; Muyan-Ozcelik, Pinar; Owens, John D

2008-08-01

The advent of readily available temporal imaging or time series volumetric (4D) imaging has become an indispensable component of treatment planning and adaptive radiotherapy (ART) at many radiotherapy centers. Deformable image registration (DIR) is also used in other areas of medical imaging, including motion corrected image reconstruction. Due to long computation time, clinical applications of DIR in radiation therapy and elsewhere have been limited and consequently relegated to offline analysis. With the recent advances in hardware and software, graphics processing unit (GPU) based computing is an emerging technology for general purpose computation, including DIR, and is suitable for highly parallelized computing. However, traditional general purpose computation on the GPU is limited because the constraints of the available programming platforms. As well, compared to CPU programming, the GPU currently has reduced dedicated processor memory, which can limit the useful working data set for parallelized processing. We present an implementation of the demons algorithm using the NVIDIA 8800 GTX GPU and the new CUDA programming language. The GPU performance will be compared with single threading and multithreading CPU implementations on an Intel dual core 2.4 GHz CPU using the C programming language. CUDA provides a C-like language programming interface, and allows for direct access to the highly parallel compute units in the GPU. Comparisons for volumetric clinical lung images acquired using 4DCT were carried out. Computation time for 100 iterations in the range of 1.8-13.5 s was observed for the GPU with image size ranging from 2.0 x 10(6) to 14.2 x 10(6) pixels. The GPU registration was 55-61 times faster than the CPU for the single threading implementation, and 34-39 times faster for the multithreading implementation. For CPU based computing, the computational time generally has a linear dependence on image size for medical imaging data. Computational efficiency is characterized in terms of time per megapixels per iteration (TPMI) with units of seconds per megapixels per iteration (or spmi). For the demons algorithm, our CPU implementation yielded largely invariant values of TPMI. The mean TPMIs were 0.527 spmi and 0.335 spmi for the single threading and multithreading cases, respectively, with <2% variation over the considered image data range. For GPU computing, we achieved TPMI =0.00916 spmi with 3.7% variation, indicating optimized memory handling under CUDA. The paradigm of GPU based real-time DIR opens up a host of clinical applications for medical imaging.

Acoustic reverse-time migration using GPU card and POSIX thread based on the adaptive optimal finite-difference scheme and the hybrid absorbing boundary condition

NASA Astrophysics Data System (ADS)

Cai, Xiaohui; Liu, Yang; Ren, Zhiming

2018-06-01

Reverse-time migration (RTM) is a powerful tool for imaging geologically complex structures such as steep-dip and subsalt. However, its implementation is quite computationally expensive. Recently, as a low-cost solution, the graphic processing unit (GPU) was introduced to improve the efficiency of RTM. In the paper, we develop three ameliorative strategies to implement RTM on GPU card. First, given the high accuracy and efficiency of the adaptive optimal finite-difference (FD) method based on least squares (LS) on central processing unit (CPU), we study the optimal LS-based FD method on GPU. Second, we develop the CPU-based hybrid absorbing boundary condition (ABC) to the GPU-based one by addressing two issues of the former when introduced to GPU card: time-consuming and chaotic threads. Third, for large-scale data, the combinatorial strategy for optimal checkpointing and efficient boundary storage is introduced for the trade-off between memory and recomputation. To save the time of communication between host and disk, the portable operating system interface (POSIX) thread is utilized to create the other CPU core at the checkpoints. Applications of the three strategies on GPU with the compute unified device architecture (CUDA) programming language in RTM demonstrate their efficiency and validity.

Towards Cloud-Resolving European-Scale Climate Simulations using a fully GPU-enabled Prototype of the COSMO Regional Model

NASA Astrophysics Data System (ADS)

Leutwyler, David; Fuhrer, Oliver; Cumming, Benjamin; Lapillonne, Xavier; Gysi, Tobias; Lüthi, Daniel; Osuna, Carlos; Schär, Christoph

2014-05-01

The representation of moist convection is a major shortcoming of current global and regional climate models. State-of-the-art global models usually operate at grid spacings of 10-300 km, and therefore cannot fully resolve the relevant upscale and downscale energy cascades. Therefore parametrization of the relevant sub-grid scale processes is required. Several studies have shown that this approach entails major uncertainties for precipitation processes, which raises concerns about the model's ability to represent precipitation statistics and associated feedback processes, as well as their sensitivities to large-scale conditions. Further refining the model resolution to the kilometer scale allows representing these processes much closer to first principles and thus should yield an improved representation of the water cycle including the drivers of extreme events. Although cloud-resolving simulations are very useful tools for climate simulations and numerical weather prediction, their high horizontal resolution and consequently the small time steps needed, challenge current supercomputers to model large domains and long time scales. The recent innovations in the domain of hybrid supercomputers have led to mixed node designs with a conventional CPU and an accelerator such as a graphics processing unit (GPU). GPUs relax the necessity for cache coherency and complex memory hierarchies, but have a larger system memory-bandwidth. This is highly beneficial for low compute intensity codes such as atmospheric stencil-based models. However, to efficiently exploit these hybrid architectures, climate models need to be ported and/or redesigned. Within the framework of the Swiss High Performance High Productivity Computing initiative (HP2C) a project to port the COSMO model to hybrid architectures has recently come to and end. The product of these efforts is a version of COSMO with an improved performance on traditional x86-based clusters as well as hybrid architectures with GPUs. We present our redesign and porting approach as well as our experience and lessons learned. Furthermore, we discuss relevant performance benchmarks obtained on the new hybrid Cray XC30 system "Piz Daint" installed at the Swiss National Supercomputing Centre (CSCS), both in terms of time-to-solution as well as energy consumption. We will demonstrate a first set of short cloud-resolving climate simulations at the European-scale using the GPU-enabled COSMO prototype and elaborate our future plans on how to exploit this new model capability.

Fast MPEG-CDVS Encoder With GPU-CPU Hybrid Computing

NASA Astrophysics Data System (ADS)

Duan, Ling-Yu; Sun, Wei; Zhang, Xinfeng; Wang, Shiqi; Chen, Jie; Yin, Jianxiong; See, Simon; Huang, Tiejun; Kot, Alex C.; Gao, Wen

2018-05-01

The compact descriptors for visual search (CDVS) standard from ISO/IEC moving pictures experts group (MPEG) has succeeded in enabling the interoperability for efficient and effective image retrieval by standardizing the bitstream syntax of compact feature descriptors. However, the intensive computation of CDVS encoder unfortunately hinders its widely deployment in industry for large-scale visual search. In this paper, we revisit the merits of low complexity design of CDVS core techniques and present a very fast CDVS encoder by leveraging the massive parallel execution resources of GPU. We elegantly shift the computation-intensive and parallel-friendly modules to the state-of-the-arts GPU platforms, in which the thread block allocation and the memory access are jointly optimized to eliminate performance loss. In addition, those operations with heavy data dependence are allocated to CPU to resolve the extra but non-necessary computation burden for GPU. Furthermore, we have demonstrated the proposed fast CDVS encoder can work well with those convolution neural network approaches which has harmoniously leveraged the advantages of GPU platforms, and yielded significant performance improvements. Comprehensive experimental results over benchmarks are evaluated, which has shown that the fast CDVS encoder using GPU-CPU hybrid computing is promising for scalable visual search.

A novel heterogeneous algorithm to simulate multiphase flow in porous media on multicore CPU-GPU systems

NASA Astrophysics Data System (ADS)

McClure, J. E.; Prins, J. F.; Miller, C. T.

2014-07-01

Multiphase flow implementations of the lattice Boltzmann method (LBM) are widely applied to the study of porous medium systems. In this work, we construct a new variant of the popular "color" LBM for two-phase flow in which a three-dimensional, 19-velocity (D3Q19) lattice is used to compute the momentum transport solution while a three-dimensional, seven velocity (D3Q7) lattice is used to compute the mass transport solution. Based on this formulation, we implement a novel heterogeneous GPU-accelerated algorithm in which the mass transport solution is computed by multiple shared memory CPU cores programmed using OpenMP while a concurrent solution of the momentum transport is performed using a GPU. The heterogeneous solution is demonstrated to provide speedup of 2.6 × as compared to multi-core CPU solution and 1.8 × compared to GPU solution due to concurrent utilization of both CPU and GPU bandwidths. Furthermore, we verify that the proposed formulation provides an accurate physical representation of multiphase flow processes and demonstrate that the approach can be applied to perform heterogeneous simulations of two-phase flow in porous media using a typical GPU-accelerated workstation.

NaNet-10: a 10GbE network interface card for the GPU-based low-level trigger of the NA62 RICH detector.

NASA Astrophysics Data System (ADS)

Ammendola, R.; Biagioni, A.; Fiorini, M.; Frezza, O.; Lonardo, A.; Lamanna, G.; Lo Cicero, F.; Martinelli, M.; Neri, I.; Paolucci, P. S.; Pastorelli, E.; Piandani, R.; Pontisso, L.; Rossetti, D.; Simula, F.; Sozzi, M.; Tosoratto, L.; Vicini, P.

2016-03-01

A GPU-based low level (L0) trigger is currently integrated in the experimental setup of the RICH detector of the NA62 experiment to assess the feasibility of building more refined physics-related trigger primitives and thus improve the trigger discriminating power. To ensure the real-time operation of the system, a dedicated data transport mechanism has been implemented: an FPGA-based Network Interface Card (NaNet-10) receives data from detectors and forwards them with low, predictable latency to the memory of the GPU performing the trigger algorithms. Results of the ring-shaped hit patterns reconstruction will be reported and discussed.

Bridging FPGA and GPU technologies for AO real-time control

NASA Astrophysics Data System (ADS)

Perret, Denis; Lainé, Maxime; Bernard, Julien; Gratadour, Damien; Sevin, Arnaud

2016-07-01

Our team has developed a common environment for high performance simulations and real-time control of AO systems based on the use of Graphics Processors Units in the context of the COMPASS project. Such a solution, based on the ability of the real time core in the simulation to provide adequate computing performance, limits the cost of developing AO RTC systems and makes them more scalable. A code developed and validated in the context of the simulation may be injected directly into the system and tested on sky. Furthermore, the use of relatively low cost components also offers significant advantages for the system hardware platform. However, the use of GPUs in an AO loop comes with drawbacks: the traditional way of offloading computation from CPU to GPUs - involving multiple copies and unacceptable overhead in kernel launching - is not well suited in a real time context. This last application requires the implementation of a solution enabling direct memory access (DMA) to the GPU memory from a third party device, bypassing the operating system. This allows this device to communicate directly with the real-time core of the simulation feeding it with the WFS camera pixel stream. We show that DMA between a custom FPGA-based frame-grabber and a computation unit (GPU, FPGA, or Coprocessor such as Xeon-phi) across PCIe allows us to get latencies compatible with what will be needed on ELTs. As a fine-grained synchronization mechanism is not yet made available by GPU vendors, we propose the use of memory polling to avoid interrupts handling and involvement of a CPU. Network and Vision protocols are handled by the FPGA-based Network Interface Card (NIC). We present the results we obtained on a complete AO loop using camera and deformable mirror simulators.

Best bang for your buck: GPU nodes for GROMACS biomolecular simulations

PubMed Central

Páll, Szilárd; Fechner, Martin; Esztermann, Ansgar; de Groot, Bert L.; Grubmüller, Helmut

2015-01-01

The molecular dynamics simulation package GROMACS runs efficiently on a wide variety of hardware from commodity workstations to high performance computing clusters. Hardware features are well‐exploited with a combination of single instruction multiple data, multithreading, and message passing interface (MPI)‐based single program multiple data/multiple program multiple data parallelism while graphics processing units (GPUs) can be used as accelerators to compute interactions off‐loaded from the CPU. Here, we evaluate which hardware produces trajectories with GROMACS 4.6 or 5.0 in the most economical way. We have assembled and benchmarked compute nodes with various CPU/GPU combinations to identify optimal compositions in terms of raw trajectory production rate, performance‐to‐price ratio, energy efficiency, and several other criteria. Although hardware prices are naturally subject to trends and fluctuations, general tendencies are clearly visible. Adding any type of GPU significantly boosts a node's simulation performance. For inexpensive consumer‐class GPUs this improvement equally reflects in the performance‐to‐price ratio. Although memory issues in consumer‐class GPUs could pass unnoticed as these cards do not support error checking and correction memory, unreliable GPUs can be sorted out with memory checking tools. Apart from the obvious determinants for cost‐efficiency like hardware expenses and raw performance, the energy consumption of a node is a major cost factor. Over the typical hardware lifetime until replacement of a few years, the costs for electrical power and cooling can become larger than the costs of the hardware itself. Taking that into account, nodes with a well‐balanced ratio of CPU and consumer‐class GPU resources produce the maximum amount of GROMACS trajectory over their lifetime. © 2015 The Authors. Journal of Computational Chemistry Published by Wiley Periodicals, Inc. PMID:26238484

Best bang for your buck: GPU nodes for GROMACS biomolecular simulations.

PubMed

Kutzner, Carsten; Páll, Szilárd; Fechner, Martin; Esztermann, Ansgar; de Groot, Bert L; Grubmüller, Helmut

2015-10-05

The molecular dynamics simulation package GROMACS runs efficiently on a wide variety of hardware from commodity workstations to high performance computing clusters. Hardware features are well-exploited with a combination of single instruction multiple data, multithreading, and message passing interface (MPI)-based single program multiple data/multiple program multiple data parallelism while graphics processing units (GPUs) can be used as accelerators to compute interactions off-loaded from the CPU. Here, we evaluate which hardware produces trajectories with GROMACS 4.6 or 5.0 in the most economical way. We have assembled and benchmarked compute nodes with various CPU/GPU combinations to identify optimal compositions in terms of raw trajectory production rate, performance-to-price ratio, energy efficiency, and several other criteria. Although hardware prices are naturally subject to trends and fluctuations, general tendencies are clearly visible. Adding any type of GPU significantly boosts a node's simulation performance. For inexpensive consumer-class GPUs this improvement equally reflects in the performance-to-price ratio. Although memory issues in consumer-class GPUs could pass unnoticed as these cards do not support error checking and correction memory, unreliable GPUs can be sorted out with memory checking tools. Apart from the obvious determinants for cost-efficiency like hardware expenses and raw performance, the energy consumption of a node is a major cost factor. Over the typical hardware lifetime until replacement of a few years, the costs for electrical power and cooling can become larger than the costs of the hardware itself. Taking that into account, nodes with a well-balanced ratio of CPU and consumer-class GPU resources produce the maximum amount of GROMACS trajectory over their lifetime. © 2015 The Authors. Journal of Computational Chemistry Published by Wiley Periodicals, Inc.

Fast, multi-channel real-time processing of signals with microsecond latency using graphics processing units.

PubMed

Rath, N; Kato, S; Levesque, J P; Mauel, M E; Navratil, G A; Peng, Q

2014-04-01

Fast, digital signal processing (DSP) has many applications. Typical hardware options for performing DSP are field-programmable gate arrays (FPGAs), application-specific integrated DSP chips, or general purpose personal computer systems. This paper presents a novel DSP platform that has been developed for feedback control on the HBT-EP tokamak device. The system runs all signal processing exclusively on a Graphics Processing Unit (GPU) to achieve real-time performance with latencies below 8 μs. Signals are transferred into and out of the GPU using PCI Express peer-to-peer direct-memory-access transfers without involvement of the central processing unit or host memory. Tests were performed on the feedback control system of the HBT-EP tokamak using forty 16-bit floating point inputs and outputs each and a sampling rate of up to 250 kHz. Signals were digitized by a D-TACQ ACQ196 module, processing done on an NVIDIA GTX 580 GPU programmed in CUDA, and analog output was generated by D-TACQ AO32CPCI modules.

Development of High-speed Visualization System of Hypocenter Data Using CUDA-based GPU computing

NASA Astrophysics Data System (ADS)

Kumagai, T.; Okubo, K.; Uchida, N.; Matsuzawa, T.; Kawada, N.; Takeuchi, N.

2014-12-01

After the Great East Japan Earthquake on March 11, 2011, intelligent visualization of seismic information is becoming important to understand the earthquake phenomena. On the other hand, to date, the quantity of seismic data becomes enormous as a progress of high accuracy observation network; we need to treat many parameters (e.g., positional information, origin time, magnitude, etc.) to efficiently display the seismic information. Therefore, high-speed processing of data and image information is necessary to handle enormous amounts of seismic data. Recently, GPU (Graphic Processing Unit) is used as an acceleration tool for data processing and calculation in various study fields. This movement is called GPGPU (General Purpose computing on GPUs). In the last few years the performance of GPU keeps on improving rapidly. GPU computing gives us the high-performance computing environment at a lower cost than before. Moreover, use of GPU has an advantage of visualization of processed data, because GPU is originally architecture for graphics processing. In the GPU computing, the processed data is always stored in the video memory. Therefore, we can directly write drawing information to the VRAM on the video card by combining CUDA and the graphics API. In this study, we employ CUDA and OpenGL and/or DirectX to realize full-GPU implementation. This method makes it possible to write drawing information to the VRAM on the video card without PCIe bus data transfer: It enables the high-speed processing of seismic data. The present study examines the GPU computing-based high-speed visualization and the feasibility for high-speed visualization system of hypocenter data.

A Fully GPU-Based Ray-Driven Backprojector via a Ray-Culling Scheme with Voxel-Level Parallelization for Cone-Beam CT Reconstruction.

PubMed

Park, Hyeong-Gyu; Shin, Yeong-Gil; Lee, Ho

2015-12-01

A ray-driven backprojector is based on ray-tracing, which computes the length of the intersection between the ray paths and each voxel to be reconstructed. To reduce the computational burden caused by these exhaustive intersection tests, we propose a fully graphics processing unit (GPU)-based ray-driven backprojector in conjunction with a ray-culling scheme that enables straightforward parallelization without compromising the high computing performance of a GPU. The purpose of the ray-culling scheme is to reduce the number of ray-voxel intersection tests by excluding rays irrelevant to a specific voxel computation. This rejection step is based on an axis-aligned bounding box (AABB) enclosing a region of voxel projection, where eight vertices of each voxel are projected onto the detector plane. The range of the rectangular-shaped AABB is determined by min/max operations on the coordinates in the region. Using the indices of pixels inside the AABB, the rays passing through the voxel can be identified and the voxel is weighted as the length of intersection between the voxel and the ray. This procedure makes it possible to reflect voxel-level parallelization, allowing an independent calculation at each voxel, which is feasible for a GPU implementation. To eliminate redundant calculations during ray-culling, a shared-memory optimization is applied to exploit the GPU memory hierarchy. In experimental results using real measurement data with phantoms, the proposed GPU-based ray-culling scheme reconstructed a volume of resolution 28032803176 in 77 seconds from 680 projections of resolution 10243768 , which is 26 times and 7.5 times faster than standard CPU-based and GPU-based ray-driven backprojectors, respectively. Qualitative and quantitative analyses showed that the ray-driven backprojector provides high-quality reconstruction images when compared with those generated by the Feldkamp-Davis-Kress algorithm using a pixel-driven backprojector, with an average of 2.5 times higher contrast-to-noise ratio, 1.04 times higher universal quality index, and 1.39 times higher normalized mutual information. © The Author(s) 2014.

TU-EF-304-07: Monte Carlo-Based Inverse Treatment Plan Optimization for Intensity Modulated Proton Therapy

DOE Office of Scientific and Technical Information (OSTI.GOV)

Li, Y; UT Southwestern Medical Center, Dallas, TX; Tian, Z

2015-06-15

Purpose: Intensity-modulated proton therapy (IMPT) is increasingly used in proton therapy. For IMPT optimization, Monte Carlo (MC) is desired for spots dose calculations because of its high accuracy, especially in cases with a high level of heterogeneity. It is also preferred in biological optimization problems due to the capability of computing quantities related to biological effects. However, MC simulation is typically too slow to be used for this purpose. Although GPU-based MC engines have become available, the achieved efficiency is still not ideal. The purpose of this work is to develop a new optimization scheme to include GPU-based MC intomore » IMPT. Methods: A conventional approach using MC in IMPT simply calls the MC dose engine repeatedly for each spot dose calculations. However, this is not the optimal approach, because of the unnecessary computations on some spots that turned out to have very small weights after solving the optimization problem. GPU-memory writing conflict occurring at a small beam size also reduces computational efficiency. To solve these problems, we developed a new framework that iteratively performs MC dose calculations and plan optimizations. At each dose calculation step, the particles were sampled from different spots altogether with Metropolis algorithm, such that the particle number is proportional to the latest optimized spot intensity. Simultaneously transporting particles from multiple spots also mitigated the memory writing conflict problem. Results: We have validated the proposed MC-based optimization schemes in one prostate case. The total computation time of our method was ∼5–6 min on one NVIDIA GPU card, including both spot dose calculation and plan optimization, whereas a conventional method naively using the same GPU-based MC engine were ∼3 times slower. Conclusion: A fast GPU-based MC dose calculation method along with a novel optimization workflow is developed. The high efficiency makes it attractive for clinical usages.« less

Modeling parameterized geometry in GPU-based Monte Carlo particle transport simulation for radiotherapy.

PubMed

Chi, Yujie; Tian, Zhen; Jia, Xun

2016-08-07

Monte Carlo (MC) particle transport simulation on a graphics-processing unit (GPU) platform has been extensively studied recently due to the efficiency advantage achieved via massive parallelization. Almost all of the existing GPU-based MC packages were developed for voxelized geometry. This limited application scope of these packages. The purpose of this paper is to develop a module to model parametric geometry and integrate it in GPU-based MC simulations. In our module, each continuous region was defined by its bounding surfaces that were parameterized by quadratic functions. Particle navigation functions in this geometry were developed. The module was incorporated to two previously developed GPU-based MC packages and was tested in two example problems: (1) low energy photon transport simulation in a brachytherapy case with a shielded cylinder applicator and (2) MeV coupled photon/electron transport simulation in a phantom containing several inserts of different shapes. In both cases, the calculated dose distributions agreed well with those calculated in the corresponding voxelized geometry. The averaged dose differences were 1.03% and 0.29%, respectively. We also used the developed package to perform simulations of a Varian VS 2000 brachytherapy source and generated a phase-space file. The computation time under the parameterized geometry depended on the memory location storing the geometry data. When the data was stored in GPU's shared memory, the highest computational speed was achieved. Incorporation of parameterized geometry yielded a computation time that was ~3 times of that in the corresponding voxelized geometry. We also developed a strategy to use an auxiliary index array to reduce frequency of geometry calculations and hence improve efficiency. With this strategy, the computational time ranged in 1.75-2.03 times of the voxelized geometry for coupled photon/electron transport depending on the voxel dimension of the auxiliary index array, and in 0.69-1.23 times for photon only transport.

Adaptation of a Multi-Block Structured Solver for Effective Use in a Hybrid CPU/GPU Massively Parallel Environment

NASA Astrophysics Data System (ADS)

Gutzwiller, David; Gontier, Mathieu; Demeulenaere, Alain

2014-11-01

Multi-Block structured solvers hold many advantages over their unstructured counterparts, such as a smaller memory footprint and efficient serial performance. Historically, multi-block structured solvers have not been easily adapted for use in a High Performance Computing (HPC) environment, and the recent trend towards hybrid GPU/CPU architectures has further complicated the situation. This paper will elaborate on developments and innovations applied to the NUMECA FINE/Turbo solver that have allowed near-linear scalability with real-world problems on over 250 hybrid GPU/GPU cluster nodes. Discussion will focus on the implementation of virtual partitioning and load balancing algorithms using a novel meta-block concept. This implementation is transparent to the user, allowing all pre- and post-processing steps to be performed using a simple, unpartitioned grid topology. Additional discussion will elaborate on developments that have improved parallel performance, including fully parallel I/O with the ADIOS API and the GPU porting of the computationally heavy CPUBooster convergence acceleration module. Head of HPC and Release Management, Numeca International.

General purpose molecular dynamics simulations fully implemented on graphics processing units

NASA Astrophysics Data System (ADS)

Anderson, Joshua A.; Lorenz, Chris D.; Travesset, A.

2008-05-01

Graphics processing units (GPUs), originally developed for rendering real-time effects in computer games, now provide unprecedented computational power for scientific applications. In this paper, we develop a general purpose molecular dynamics code that runs entirely on a single GPU. It is shown that our GPU implementation provides a performance equivalent to that of fast 30 processor core distributed memory cluster. Our results show that GPUs already provide an inexpensive alternative to such clusters and discuss implications for the future.

Simulating spin models on GPU

NASA Astrophysics Data System (ADS)

Weigel, Martin

2011-09-01

Over the last couple of years it has been realized that the vast computational power of graphics processing units (GPUs) could be harvested for purposes other than the video game industry. This power, which at least nominally exceeds that of current CPUs by large factors, results from the relative simplicity of the GPU architectures as compared to CPUs, combined with a large number of parallel processing units on a single chip. To benefit from this setup for general computing purposes, the problems at hand need to be prepared in a way to profit from the inherent parallelism and hierarchical structure of memory accesses. In this contribution I discuss the performance potential for simulating spin models, such as the Ising model, on GPU as compared to conventional simulations on CPU.

Fast MPEG-CDVS Encoder With GPU-CPU Hybrid Computing.

PubMed

Duan, Ling-Yu; Sun, Wei; Zhang, Xinfeng; Wang, Shiqi; Chen, Jie; Yin, Jianxiong; See, Simon; Huang, Tiejun; Kot, Alex C; Gao, Wen

2018-05-01

The compact descriptors for visual search (CDVS) standard from ISO/IEC moving pictures experts group has succeeded in enabling the interoperability for efficient and effective image retrieval by standardizing the bitstream syntax of compact feature descriptors. However, the intensive computation of a CDVS encoder unfortunately hinders its widely deployment in industry for large-scale visual search. In this paper, we revisit the merits of low complexity design of CDVS core techniques and present a very fast CDVS encoder by leveraging the massive parallel execution resources of graphics processing unit (GPU). We elegantly shift the computation-intensive and parallel-friendly modules to the state-of-the-arts GPU platforms, in which the thread block allocation as well as the memory access mechanism are jointly optimized to eliminate performance loss. In addition, those operations with heavy data dependence are allocated to CPU for resolving the extra but non-necessary computation burden for GPU. Furthermore, we have demonstrated the proposed fast CDVS encoder can work well with those convolution neural network approaches which enables to leverage the advantages of GPU platforms harmoniously, and yield significant performance improvements. Comprehensive experimental results over benchmarks are evaluated, which has shown that the fast CDVS encoder using GPU-CPU hybrid computing is promising for scalable visual search.

Incompressible SPH (ISPH) with fast Poisson solver on a GPU

NASA Astrophysics Data System (ADS)

Chow, Alex D.; Rogers, Benedict D.; Lind, Steven J.; Stansby, Peter K.

2018-05-01

This paper presents a fast incompressible SPH (ISPH) solver implemented to run entirely on a graphics processing unit (GPU) capable of simulating several millions of particles in three dimensions on a single GPU. The ISPH algorithm is implemented by converting the highly optimised open-source weakly-compressible SPH (WCSPH) code DualSPHysics to run ISPH on the GPU, combining it with the open-source linear algebra library ViennaCL for fast solutions of the pressure Poisson equation (PPE). Several challenges are addressed with this research: constructing a PPE matrix every timestep on the GPU for moving particles, optimising the limited GPU memory, and exploiting fast matrix solvers. The ISPH pressure projection algorithm is implemented as 4 separate stages, each with a particle sweep, including an algorithm for the population of the PPE matrix suitable for the GPU, and mixed precision storage methods. An accurate and robust ISPH boundary condition ideal for parallel processing is also established by adapting an existing WCSPH boundary condition for ISPH. A variety of validation cases are presented: an impulsively started plate, incompressible flow around a moving square in a box, and dambreaks (2-D and 3-D) which demonstrate the accuracy, flexibility, and speed of the methodology. Fragmentation of the free surface is shown to influence the performance of matrix preconditioners and therefore the PPE matrix solution time. The Jacobi preconditioner demonstrates robustness and reliability in the presence of fragmented flows. For a dambreak simulation, GPU speed ups demonstrate up to 10-18 times and 1.1-4.5 times compared to single-threaded and 16-threaded CPU run times respectively.

Multi-GPU implementation of a VMAT treatment plan optimization algorithm

DOE Office of Scientific and Technical Information (OSTI.GOV)

Tian, Zhen, E-mail: Zhen.Tian@UTSouthwestern.edu, E-mail: Xun.Jia@UTSouthwestern.edu, E-mail: Steve.Jiang@UTSouthwestern.edu; Folkerts, Michael; Tan, Jun

Purpose: Volumetric modulated arc therapy (VMAT) optimization is a computationally challenging problem due to its large data size, high degrees of freedom, and many hardware constraints. High-performance graphics processing units (GPUs) have been used to speed up the computations. However, GPU’s relatively small memory size cannot handle cases with a large dose-deposition coefficient (DDC) matrix in cases of, e.g., those with a large target size, multiple targets, multiple arcs, and/or small beamlet size. The main purpose of this paper is to report an implementation of a column-generation-based VMAT algorithm, previously developed in the authors’ group, on a multi-GPU platform tomore » solve the memory limitation problem. While the column-generation-based VMAT algorithm has been previously developed, the GPU implementation details have not been reported. Hence, another purpose is to present detailed techniques employed for GPU implementation. The authors also would like to utilize this particular problem as an example problem to study the feasibility of using a multi-GPU platform to solve large-scale problems in medical physics. Methods: The column-generation approach generates VMAT apertures sequentially by solving a pricing problem (PP) and a master problem (MP) iteratively. In the authors’ method, the sparse DDC matrix is first stored on a CPU in coordinate list format (COO). On the GPU side, this matrix is split into four submatrices according to beam angles, which are stored on four GPUs in compressed sparse row format. Computation of beamlet price, the first step in PP, is accomplished using multi-GPUs. A fast inter-GPU data transfer scheme is accomplished using peer-to-peer access. The remaining steps of PP and MP problems are implemented on CPU or a single GPU due to their modest problem scale and computational loads. Barzilai and Borwein algorithm with a subspace step scheme is adopted here to solve the MP problem. A head and neck (H and N) cancer case is then used to validate the authors’ method. The authors also compare their multi-GPU implementation with three different single GPU implementation strategies, i.e., truncating DDC matrix (S1), repeatedly transferring DDC matrix between CPU and GPU (S2), and porting computations involving DDC matrix to CPU (S3), in terms of both plan quality and computational efficiency. Two more H and N patient cases and three prostate cases are used to demonstrate the advantages of the authors’ method. Results: The authors’ multi-GPU implementation can finish the optimization process within ∼1 min for the H and N patient case. S1 leads to an inferior plan quality although its total time was 10 s shorter than the multi-GPU implementation due to the reduced matrix size. S2 and S3 yield the same plan quality as the multi-GPU implementation but take ∼4 and ∼6 min, respectively. High computational efficiency was consistently achieved for the other five patient cases tested, with VMAT plans of clinically acceptable quality obtained within 23–46 s. Conversely, to obtain clinically comparable or acceptable plans for all six of these VMAT cases that the authors have tested in this paper, the optimization time needed in a commercial TPS system on CPU was found to be in an order of several minutes. Conclusions: The results demonstrate that the multi-GPU implementation of the authors’ column-generation-based VMAT optimization can handle the large-scale VMAT optimization problem efficiently without sacrificing plan quality. The authors’ study may serve as an example to shed some light on other large-scale medical physics problems that require multi-GPU techniques.« less

Accelerating adaptive inverse distance weighting interpolation algorithm on a graphics processing unit

PubMed Central

Xu, Liangliang; Xu, Nengxiong

2017-01-01

This paper focuses on designing and implementing parallel adaptive inverse distance weighting (AIDW) interpolation algorithms by using the graphics processing unit (GPU). The AIDW is an improved version of the standard IDW, which can adaptively determine the power parameter according to the data points’ spatial distribution pattern and achieve more accurate predictions than those predicted by IDW. In this paper, we first present two versions of the GPU-accelerated AIDW, i.e. the naive version without profiting from the shared memory and the tiled version taking advantage of the shared memory. We also implement the naive version and the tiled version using two data layouts, structure of arrays and array of aligned structures, on both single and double precision. We then evaluate the performance of parallel AIDW by comparing it with its corresponding serial algorithm on three different machines equipped with the GPUs GT730M, M5000 and K40c. The experimental results indicate that: (i) there is no significant difference in the computational efficiency when different data layouts are employed; (ii) the tiled version is always slightly faster than the naive version; and (iii) on single precision the achieved speed-up can be up to 763 (on the GPU M5000), while on double precision the obtained highest speed-up is 197 (on the GPU K40c). To benefit the community, all source code and testing data related to the presented parallel AIDW algorithm are publicly available. PMID:28989754

Accelerating adaptive inverse distance weighting interpolation algorithm on a graphics processing unit.

PubMed

Mei, Gang; Xu, Liangliang; Xu, Nengxiong

2017-09-01

This paper focuses on designing and implementing parallel adaptive inverse distance weighting (AIDW) interpolation algorithms by using the graphics processing unit (GPU). The AIDW is an improved version of the standard IDW, which can adaptively determine the power parameter according to the data points' spatial distribution pattern and achieve more accurate predictions than those predicted by IDW. In this paper, we first present two versions of the GPU-accelerated AIDW, i.e. the naive version without profiting from the shared memory and the tiled version taking advantage of the shared memory. We also implement the naive version and the tiled version using two data layouts, structure of arrays and array of aligned structures, on both single and double precision. We then evaluate the performance of parallel AIDW by comparing it with its corresponding serial algorithm on three different machines equipped with the GPUs GT730M, M5000 and K40c. The experimental results indicate that: (i) there is no significant difference in the computational efficiency when different data layouts are employed; (ii) the tiled version is always slightly faster than the naive version; and (iii) on single precision the achieved speed-up can be up to 763 (on the GPU M5000), while on double precision the obtained highest speed-up is 197 (on the GPU K40c). To benefit the community, all source code and testing data related to the presented parallel AIDW algorithm are publicly available.

Particle-in-cell simulations with charge-conserving current deposition on graphic processing units

NASA Astrophysics Data System (ADS)

Ren, Chuang; Kong, Xianglong; Huang, Michael; Decyk, Viktor; Mori, Warren

2011-10-01

Recently using CUDA, we have developed an electromagnetic Particle-in-Cell (PIC) code with charge-conserving current deposition for Nvidia graphic processing units (GPU's) (Kong et al., Journal of Computational Physics 230, 1676 (2011). On a Tesla M2050 (Fermi) card, the GPU PIC code can achieve a one-particle-step process time of 1.2 - 3.2 ns in 2D and 2.3 - 7.2 ns in 3D, depending on plasma temperatures. In this talk we will discuss novel algorithms for GPU-PIC including charge-conserving current deposition scheme with few branching and parallel particle sorting. These algorithms have made efficient use of the GPU shared memory. We will also discuss how to replace the computation kernels of existing parallel CPU codes while keeping their parallel structures. This work was supported by U.S. Department of Energy under Grant Nos. DE-FG02-06ER54879 and DE-FC02-04ER54789 and by NSF under Grant Nos. PHY-0903797 and CCF-0747324.

GPU computing of compressible flow problems by a meshless method with space-filling curves

NASA Astrophysics Data System (ADS)

Ma, Z. H.; Wang, H.; Pu, S. H.

2014-04-01

A graphic processing unit (GPU) implementation of a meshless method for solving compressible flow problems is presented in this paper. Least-square fit is used to discretize the spatial derivatives of Euler equations and an upwind scheme is applied to estimate the flux terms. The compute unified device architecture (CUDA) C programming model is employed to efficiently and flexibly port the meshless solver from CPU to GPU. Considering the data locality of randomly distributed points, space-filling curves are adopted to re-number the points in order to improve the memory performance. Detailed evaluations are firstly carried out to assess the accuracy and conservation property of the underlying numerical method. Then the GPU accelerated flow solver is used to solve external steady flows over aerodynamic configurations. Representative results are validated through extensive comparisons with the experimental, finite volume or other available reference solutions. Performance analysis reveals that the running time cost of simulations is significantly reduced while impressive (more than an order of magnitude) speedups are achieved.

The density matrix renormalization group algorithm on kilo-processor architectures: Implementation and trade-offs

NASA Astrophysics Data System (ADS)

Nemes, Csaba; Barcza, Gergely; Nagy, Zoltán; Legeza, Örs; Szolgay, Péter

2014-06-01

In the numerical analysis of strongly correlated quantum lattice models one of the leading algorithms developed to balance the size of the effective Hilbert space and the accuracy of the simulation is the density matrix renormalization group (DMRG) algorithm, in which the run-time is dominated by the iterative diagonalization of the Hamilton operator. As the most time-dominant step of the diagonalization can be expressed as a list of dense matrix operations, the DMRG is an appealing candidate to fully utilize the computing power residing in novel kilo-processor architectures. In the paper a smart hybrid CPU-GPU implementation is presented, which exploits the power of both CPU and GPU and tolerates problems exceeding the GPU memory size. Furthermore, a new CUDA kernel has been designed for asymmetric matrix-vector multiplication to accelerate the rest of the diagonalization. Besides the evaluation of the GPU implementation, the practical limits of an FPGA implementation are also discussed.

Accelerating a three-dimensional eco-hydrological cellular automaton on GPGPU with OpenCL

NASA Astrophysics Data System (ADS)

Senatore, Alfonso; D'Ambrosio, Donato; De Rango, Alessio; Rongo, Rocco; Spataro, William; Straface, Salvatore; Mendicino, Giuseppe

2016-10-01

This work presents an effective implementation of a numerical model for complete eco-hydrological Cellular Automata modeling on Graphical Processing Units (GPU) with OpenCL (Open Computing Language) for heterogeneous computation (i.e., on CPUs and/or GPUs). Different types of parallel implementations were carried out (e.g., use of fast local memory, loop unrolling, etc), showing increasing performance improvements in terms of speedup, adopting also some original optimizations strategies. Moreover, numerical analysis of results (i.e., comparison of CPU and GPU outcomes in terms of rounding errors) have proven to be satisfactory. Experiments were carried out on a workstation with two CPUs (Intel Xeon E5440 at 2.83GHz), one GPU AMD R9 280X and one GPU nVIDIA Tesla K20c. Results have been extremely positive, but further testing should be performed to assess the functionality of the adopted strategies on other complete models and their ability to fruitfully exploit parallel systems resources.

GPU-based parallel algorithm for blind image restoration using midfrequency-based methods

NASA Astrophysics Data System (ADS)

Xie, Lang; Luo, Yi-han; Bao, Qi-liang

2013-08-01

GPU-based general-purpose computing is a new branch of modern parallel computing, so the study of parallel algorithms specially designed for GPU hardware architecture is of great significance. In order to solve the problem of high computational complexity and poor real-time performance in blind image restoration, the midfrequency-based algorithm for blind image restoration was analyzed and improved in this paper. Furthermore, a midfrequency-based filtering method is also used to restore the image hardly with any recursion or iteration. Combining the algorithm with data intensiveness, data parallel computing and GPU execution model of single instruction and multiple threads, a new parallel midfrequency-based algorithm for blind image restoration is proposed in this paper, which is suitable for stream computing of GPU. In this algorithm, the GPU is utilized to accelerate the estimation of class-G point spread functions and midfrequency-based filtering. Aiming at better management of the GPU threads, the threads in a grid are scheduled according to the decomposition of the filtering data in frequency domain after the optimization of data access and the communication between the host and the device. The kernel parallelism structure is determined by the decomposition of the filtering data to ensure the transmission rate to get around the memory bandwidth limitation. The results show that, with the new algorithm, the operational speed is significantly increased and the real-time performance of image restoration is effectively improved, especially for high-resolution images.

GPU implementation of the simplex identification via split augmented Lagrangian

NASA Astrophysics Data System (ADS)

Sevilla, Jorge; Nascimento, José M. P.

2015-10-01

Hyperspectral imaging can be used for object detection and for discriminating between different objects based on their spectral characteristics. One of the main problems of hyperspectral data analysis is the presence of mixed pixels, due to the low spatial resolution of such images. This means that several spectrally pure signatures (endmembers) are combined into the same mixed pixel. Linear spectral unmixing follows an unsupervised approach which aims at inferring pure spectral signatures and their material fractions at each pixel of the scene. The huge data volumes acquired by such sensors put stringent requirements on processing and unmixing methods. This paper proposes an efficient implementation of a unsupervised linear unmixing method on GPUs using CUDA. The method finds the smallest simplex by solving a sequence of nonsmooth convex subproblems using variable splitting to obtain a constraint formulation, and then applying an augmented Lagrangian technique. The parallel implementation of SISAL presented in this work exploits the GPU architecture at low level, using shared memory and coalesced accesses to memory. The results herein presented indicate that the GPU implementation can significantly accelerate the method's execution over big datasets while maintaining the methods accuracy.

GPU-based stochastic-gradient optimization for non-rigid medical image registration in time-critical applications

NASA Astrophysics Data System (ADS)

Bhosale, Parag; Staring, Marius; Al-Ars, Zaid; Berendsen, Floris F.

2018-03-01

Currently, non-rigid image registration algorithms are too computationally intensive to use in time-critical applications. Existing implementations that focus on speed typically address this by either parallelization on GPU-hardware, or by introducing methodically novel techniques into CPU-oriented algorithms. Stochastic gradient descent (SGD) optimization and variations thereof have proven to drastically reduce the computational burden for CPU-based image registration, but have not been successfully applied in GPU hardware due to its stochastic nature. This paper proposes 1) NiftyRegSGD, a SGD optimization for the GPU-based image registration tool NiftyReg, 2) random chunk sampler, a new random sampling strategy that better utilizes the memory bandwidth of GPU hardware. Experiments have been performed on 3D lung CT data of 19 patients, which compared NiftyRegSGD (with and without random chunk sampler) with CPU-based elastix Fast Adaptive SGD (FASGD) and NiftyReg. The registration runtime was 21.5s, 4.4s and 2.8s for elastix-FASGD, NiftyRegSGD without, and NiftyRegSGD with random chunk sampling, respectively, while similar accuracy was obtained. Our method is publicly available at https://github.com/SuperElastix/NiftyRegSGD.

Streaming parallel GPU acceleration of large-scale filter-based spiking neural networks.

PubMed

Slażyński, Leszek; Bohte, Sander

2012-01-01

The arrival of graphics processing (GPU) cards suitable for massively parallel computing promises affordable large-scale neural network simulation previously only available at supercomputing facilities. While the raw numbers suggest that GPUs may outperform CPUs by at least an order of magnitude, the challenge is to develop fine-grained parallel algorithms to fully exploit the particulars of GPUs. Computation in a neural network is inherently parallel and thus a natural match for GPU architectures: given inputs, the internal state for each neuron can be updated in parallel. We show that for filter-based spiking neurons, like the Spike Response Model, the additive nature of membrane potential dynamics enables additional update parallelism. This also reduces the accumulation of numerical errors when using single precision computation, the native precision of GPUs. We further show that optimizing simulation algorithms and data structures to the GPU's architecture has a large pay-off: for example, matching iterative neural updating to the memory architecture of the GPU speeds up this simulation step by a factor of three to five. With such optimizations, we can simulate in better-than-realtime plausible spiking neural networks of up to 50 000 neurons, processing over 35 million spiking events per second.

GPU and APU computations of Finite Time Lyapunov Exponent fields

NASA Astrophysics Data System (ADS)

Conti, Christian; Rossinelli, Diego; Koumoutsakos, Petros

2012-03-01

We present GPU and APU accelerated computations of Finite-Time Lyapunov Exponent (FTLE) fields. The calculation of FTLEs is a computationally intensive process, as in order to obtain the sharp ridges associated with the Lagrangian Coherent Structures an extensive resampling of the flow field is required. The computational performance of this resampling is limited by the memory bandwidth of the underlying computer architecture. The present technique harnesses data-parallel execution of many-core architectures and relies on fast and accurate evaluations of moment conserving functions for the mesh to particle interpolations. We demonstrate how the computation of FTLEs can be efficiently performed on a GPU and on an APU through OpenCL and we report over one order of magnitude improvements over multi-threaded executions in FTLE computations of bluff body flows.

Mendel-GPU: haplotyping and genotype imputation on graphics processing units

PubMed Central

Chen, Gary K.; Wang, Kai; Stram, Alex H.; Sobel, Eric M.; Lange, Kenneth

2012-01-01

Motivation: In modern sequencing studies, one can improve the confidence of genotype calls by phasing haplotypes using information from an external reference panel of fully typed unrelated individuals. However, the computational demands are so high that they prohibit researchers with limited computational resources from haplotyping large-scale sequence data. Results: Our graphics processing unit based software delivers haplotyping and imputation accuracies comparable to competing programs at a fraction of the computational cost and peak memory demand. Availability: Mendel-GPU, our OpenCL software, runs on Linux platforms and is portable across AMD and nVidia GPUs. Users can download both code and documentation at http://code.google.com/p/mendel-gpu/. Contact: gary.k.chen@usc.edu Supplementary information: Supplementary data are available at Bioinformatics online. PMID:22954633

DOE Office of Scientific and Technical Information (OSTI.GOV)

Edwards, Harold C.; Ibanez, Daniel Alejandro

This report documents the ASC/ATDM Kokkos deliverable "Production Portable Dy- namic Task DAG Capability." This capability enables applications to create and execute a dynamic task DAG ; a collection of heterogeneous computational tasks with a directed acyclic graph (DAG) of "execute after" dependencies where tasks and their dependencies are dynamically created and destroyed as tasks execute. The Kokkos task scheduler executes the dynamic task DAG on the target execution resource; e.g. a multicore CPU, a manycore CPU such as Intel's Knights Landing (KNL), or an NVIDIA GPU. Several major technical challenges had to be addressed during development of Kokkos' Taskmore » DAG capability: (1) portability to a GPU with it's simplified hardware and micro- runtime, (2) thread-scalable memory allocation and deallocation from a bounded pool of memory, (3) thread-scalable scheduler for dynamic task DAG, (4) usability by applications.« less

Pyglidein - A Simple HTCondor Glidein Service

NASA Astrophysics Data System (ADS)

Schultz, D.; Riedel, B.; Merino, G.

2017-10-01

A major challenge for data processing and analysis at the IceCube Neutrino Observatory presents itself in connecting a large set of individual clusters together to form a computing grid. Most of these clusters do not provide a “standard” grid interface. Using a local account on each submit machine, HTCondor glideins can be submitted to virtually any type of scheduler. The glideins then connect back to a main HTCondor pool, where jobs can run normally with no special syntax. To respond to dynamic load, a simple server advertises the number of idle jobs in the queue and the resources they request. The submit script can query this server to optimize glideins to what is needed, or not submit if there is no demand. Configuring HTCondor dynamic slots in the glideins allows us to efficiently handle varying memory requirements as well as whole-node jobs. One step of the IceCube simulation chain, photon propagation in the ice, heavily relies on GPUs for faster execution. Therefore, one important requirement for any workload management system in IceCube is to handle GPU resources properly. Within the pyglidein system, we have successfully configured HTCondor glideins to use any GPU allocated to it, with jobs using the standard HTCondor GPU syntax to request and use a GPU. This mechanism allows us to seamlessly integrate our local GPU cluster with remote non-Grid GPU clusters, including specially allocated resources at XSEDE supercomputers.

Cpu/gpu Computing for AN Implicit Multi-Block Compressible Navier-Stokes Solver on Heterogeneous Platform

NASA Astrophysics Data System (ADS)

Deng, Liang; Bai, Hanli; Wang, Fang; Xu, Qingxin

2016-06-01

CPU/GPU computing allows scientists to tremendously accelerate their numerical codes. In this paper, we port and optimize a double precision alternating direction implicit (ADI) solver for three-dimensional compressible Navier-Stokes equations from our in-house Computational Fluid Dynamics (CFD) software on heterogeneous platform. First, we implement a full GPU version of the ADI solver to remove a lot of redundant data transfers between CPU and GPU, and then design two fine-grain schemes, namely “one-thread-one-point” and “one-thread-one-line”, to maximize the performance. Second, we present a dual-level parallelization scheme using the CPU/GPU collaborative model to exploit the computational resources of both multi-core CPUs and many-core GPUs within the heterogeneous platform. Finally, considering the fact that memory on a single node becomes inadequate when the simulation size grows, we present a tri-level hybrid programming pattern MPI-OpenMP-CUDA that merges fine-grain parallelism using OpenMP and CUDA threads with coarse-grain parallelism using MPI for inter-node communication. We also propose a strategy to overlap the computation with communication using the advanced features of CUDA and MPI programming. We obtain speedups of 6.0 for the ADI solver on one Tesla M2050 GPU in contrast to two Xeon X5670 CPUs. Scalability tests show that our implementation can offer significant performance improvement on heterogeneous platform.

OpenACC acceleration of an unstructured CFD solver based on a reconstructed discontinuous Galerkin method for compressible flows

DOE PAGES

Xia, Yidong; Lou, Jialin; Luo, Hong; ...

2015-02-09

Here, an OpenACC directive-based graphics processing unit (GPU) parallel scheme is presented for solving the compressible Navier–Stokes equations on 3D hybrid unstructured grids with a third-order reconstructed discontinuous Galerkin method. The developed scheme requires the minimum code intrusion and algorithm alteration for upgrading a legacy solver with the GPU computing capability at very little extra effort in programming, which leads to a unified and portable code development strategy. A face coloring algorithm is adopted to eliminate the memory contention because of the threading of internal and boundary face integrals. A number of flow problems are presented to verify the implementationmore » of the developed scheme. Timing measurements were obtained by running the resulting GPU code on one Nvidia Tesla K20c GPU card (Nvidia Corporation, Santa Clara, CA, USA) and compared with those obtained by running the equivalent Message Passing Interface (MPI) parallel CPU code on a compute node (consisting of two AMD Opteron 6128 eight-core CPUs (Advanced Micro Devices, Inc., Sunnyvale, CA, USA)). Speedup factors of up to 24× and 1.6× for the GPU code were achieved with respect to one and 16 CPU cores, respectively. The numerical results indicate that this OpenACC-based parallel scheme is an effective and extensible approach to port unstructured high-order CFD solvers to GPU computing.« less

Adaptive multi-GPU Exchange Monte Carlo for the 3D Random Field Ising Model

NASA Astrophysics Data System (ADS)

Navarro, Cristóbal A.; Huang, Wei; Deng, Youjin

2016-08-01

This work presents an adaptive multi-GPU Exchange Monte Carlo approach for the simulation of the 3D Random Field Ising Model (RFIM). The design is based on a two-level parallelization. The first level, spin-level parallelism, maps the parallel computation as optimal 3D thread-blocks that simulate blocks of spins in shared memory with minimal halo surface, assuming a constant block volume. The second level, replica-level parallelism, uses multi-GPU computation to handle the simulation of an ensemble of replicas. CUDA's concurrent kernel execution feature is used in order to fill the occupancy of each GPU with many replicas, providing a performance boost that is more notorious at the smallest values of L. In addition to the two-level parallel design, the work proposes an adaptive multi-GPU approach that dynamically builds a proper temperature set free of exchange bottlenecks. The strategy is based on mid-point insertions at the temperature gaps where the exchange rate is most compromised. The extra work generated by the insertions is balanced across the GPUs independently of where the mid-point insertions were performed. Performance results show that spin-level performance is approximately two orders of magnitude faster than a single-core CPU version and one order of magnitude faster than a parallel multi-core CPU version running on 16-cores. Multi-GPU performance is highly convenient under a weak scaling setting, reaching up to 99 % efficiency as long as the number of GPUs and L increase together. The combination of the adaptive approach with the parallel multi-GPU design has extended our possibilities of simulation to sizes of L = 32 , 64 for a workstation with two GPUs. Sizes beyond L = 64 can eventually be studied using larger multi-GPU systems.

GENIE: a software package for gene-gene interaction analysis in genetic association studies using multiple GPU or CPU cores.

PubMed

Chikkagoudar, Satish; Wang, Kai; Li, Mingyao

2011-05-26

Gene-gene interaction in genetic association studies is computationally intensive when a large number of SNPs are involved. Most of the latest Central Processing Units (CPUs) have multiple cores, whereas Graphics Processing Units (GPUs) also have hundreds of cores and have been recently used to implement faster scientific software. However, currently there are no genetic analysis software packages that allow users to fully utilize the computing power of these multi-core devices for genetic interaction analysis for binary traits. Here we present a novel software package GENIE, which utilizes the power of multiple GPU or CPU processor cores to parallelize the interaction analysis. GENIE reads an entire genetic association study dataset into memory and partitions the dataset into fragments with non-overlapping sets of SNPs. For each fragment, GENIE analyzes: 1) the interaction of SNPs within it in parallel, and 2) the interaction between the SNPs of the current fragment and other fragments in parallel. We tested GENIE on a large-scale candidate gene study on high-density lipoprotein cholesterol. Using an NVIDIA Tesla C1060 graphics card, the GPU mode of GENIE achieves a speedup of 27 times over its single-core CPU mode run. GENIE is open-source, economical, user-friendly, and scalable. Since the computing power and memory capacity of graphics cards are increasing rapidly while their cost is going down, we anticipate that GENIE will achieve greater speedups with faster GPU cards. Documentation, source code, and precompiled binaries can be downloaded from http://www.cceb.upenn.edu/~mli/software/GENIE/.

Multidisciplinary Simulation Acceleration using Multiple Shared-Memory Graphical Processing Units

NASA Astrophysics Data System (ADS)

Kemal, Jonathan Yashar

For purposes of optimizing and analyzing turbomachinery and other designs, the unsteady Favre-averaged flow-field differential equations for an ideal compressible gas can be solved in conjunction with the heat conduction equation. We solve all equations using the finite-volume multiple-grid numerical technique, with the dual time-step scheme used for unsteady simulations. Our numerical solver code targets CUDA-capable Graphical Processing Units (GPUs) produced by NVIDIA. Making use of MPI, our solver can run across networked compute notes, where each MPI process can use either a GPU or a Central Processing Unit (CPU) core for primary solver calculations. We use NVIDIA Tesla C2050/C2070 GPUs based on the Fermi architecture, and compare our resulting performance against Intel Zeon X5690 CPUs. Solver routines converted to CUDA typically run about 10 times faster on a GPU for sufficiently dense computational grids. We used a conjugate cylinder computational grid and ran a turbulent steady flow simulation using 4 increasingly dense computational grids. Our densest computational grid is divided into 13 blocks each containing 1033x1033 grid points, for a total of 13.87 million grid points or 1.07 million grid points per domain block. To obtain overall speedups, we compare the execution time of the solver's iteration loop, including all resource intensive GPU-related memory copies. Comparing the performance of 8 GPUs to that of 8 CPUs, we obtain an overall speedup of about 6.0 when using our densest computational grid. This amounts to an 8-GPU simulation running about 39.5 times faster than running than a single-CPU simulation.

GPU-accelerated iterative reconstruction for limited-data tomography in CBCT systems.

PubMed

de Molina, Claudia; Serrano, Estefania; Garcia-Blas, Javier; Carretero, Jesus; Desco, Manuel; Abella, Monica

2018-05-15

Standard cone-beam computed tomography (CBCT) involves the acquisition of at least 360 projections rotating through 360 degrees. Nevertheless, there are cases in which only a few projections can be taken in a limited angular span, such as during surgery, where rotation of the source-detector pair is limited to less than 180 degrees. Reconstruction of limited data with the conventional method proposed by Feldkamp, Davis and Kress (FDK) results in severe artifacts. Iterative methods may compensate for the lack of data by including additional prior information, although they imply a high computational burden and memory consumption. We present an accelerated implementation of an iterative method for CBCT following the Split Bregman formulation, which reduces computational time through GPU-accelerated kernels. The implementation enables the reconstruction of large volumes (>1024 3 pixels) using partitioning strategies in forward- and back-projection operations. We evaluated the algorithm on small-animal data for different scenarios with different numbers of projections, angular span, and projection size. Reconstruction time varied linearly with the number of projections and quadratically with projection size but remained almost unchanged with angular span. Forward- and back-projection operations represent 60% of the total computational burden. Efficient implementation using parallel processing and large-memory management strategies together with GPU kernels enables the use of advanced reconstruction approaches which are needed in limited-data scenarios. Our GPU implementation showed a significant time reduction (up to 48 ×) compared to a CPU-only implementation, resulting in a total reconstruction time from several hours to few minutes.

An Out-of-Core GPU based dimensionality reduction algorithm for Big Mass Spectrometry Data and its application in bottom-up Proteomics.

PubMed

Awan, Muaaz Gul; Saeed, Fahad

2017-08-01

Modern high resolution Mass Spectrometry instruments can generate millions of spectra in a single systems biology experiment. Each spectrum consists of thousands of peaks but only a small number of peaks actively contribute to deduction of peptides. Therefore, pre-processing of MS data to detect noisy and non-useful peaks are an active area of research. Most of the sequential noise reducing algorithms are impractical to use as a pre-processing step due to high time-complexity. In this paper, we present a GPU based dimensionality-reduction algorithm, called G-MSR, for MS2 spectra. Our proposed algorithm uses novel data structures which optimize the memory and computational operations inside GPU. These novel data structures include Binary Spectra and Quantized Indexed Spectra (QIS) . The former helps in communicating essential information between CPU and GPU using minimum amount of data while latter enables us to store and process complex 3-D data structure into a 1-D array structure while maintaining the integrity of MS data. Our proposed algorithm also takes into account the limited memory of GPUs and switches between in-core and out-of-core modes based upon the size of input data. G-MSR achieves a peak speed-up of 386x over its sequential counterpart and is shown to process over a million spectra in just 32 seconds. The code for this algorithm is available as a GPL open-source at GitHub at the following link: https://github.com/pcdslab/G-MSR.

GENIE: a software package for gene-gene interaction analysis in genetic association studies using multiple GPU or CPU cores

PubMed Central

2011-01-01

Background Gene-gene interaction in genetic association studies is computationally intensive when a large number of SNPs are involved. Most of the latest Central Processing Units (CPUs) have multiple cores, whereas Graphics Processing Units (GPUs) also have hundreds of cores and have been recently used to implement faster scientific software. However, currently there are no genetic analysis software packages that allow users to fully utilize the computing power of these multi-core devices for genetic interaction analysis for binary traits. Findings Here we present a novel software package GENIE, which utilizes the power of multiple GPU or CPU processor cores to parallelize the interaction analysis. GENIE reads an entire genetic association study dataset into memory and partitions the dataset into fragments with non-overlapping sets of SNPs. For each fragment, GENIE analyzes: 1) the interaction of SNPs within it in parallel, and 2) the interaction between the SNPs of the current fragment and other fragments in parallel. We tested GENIE on a large-scale candidate gene study on high-density lipoprotein cholesterol. Using an NVIDIA Tesla C1060 graphics card, the GPU mode of GENIE achieves a speedup of 27 times over its single-core CPU mode run. Conclusions GENIE is open-source, economical, user-friendly, and scalable. Since the computing power and memory capacity of graphics cards are increasing rapidly while their cost is going down, we anticipate that GENIE will achieve greater speedups with faster GPU cards. Documentation, source code, and precompiled binaries can be downloaded from http://www.cceb.upenn.edu/~mli/software/GENIE/. PMID:21615923

Integrative multicellular biological modeling: a case study of 3D epidermal development using GPU algorithms

PubMed Central

2010-01-01

Background Simulation of sophisticated biological models requires considerable computational power. These models typically integrate together numerous biological phenomena such as spatially-explicit heterogeneous cells, cell-cell interactions, cell-environment interactions and intracellular gene networks. The recent advent of programming for graphical processing units (GPU) opens up the possibility of developing more integrative, detailed and predictive biological models while at the same time decreasing the computational cost to simulate those models. Results We construct a 3D model of epidermal development and provide a set of GPU algorithms that executes significantly faster than sequential central processing unit (CPU) code. We provide a parallel implementation of the subcellular element method for individual cells residing in a lattice-free spatial environment. Each cell in our epidermal model includes an internal gene network, which integrates cellular interaction of Notch signaling together with environmental interaction of basement membrane adhesion, to specify cellular state and behaviors such as growth and division. We take a pedagogical approach to describing how modeling methods are efficiently implemented on the GPU including memory layout of data structures and functional decomposition. We discuss various programmatic issues and provide a set of design guidelines for GPU programming that are instructive to avoid common pitfalls as well as to extract performance from the GPU architecture. Conclusions We demonstrate that GPU algorithms represent a significant technological advance for the simulation of complex biological models. We further demonstrate with our epidermal model that the integration of multiple complex modeling methods for heterogeneous multicellular biological processes is both feasible and computationally tractable using this new technology. We hope that the provided algorithms and source code will be a starting point for modelers to develop their own GPU implementations, and encourage others to implement their modeling methods on the GPU and to make that code available to the wider community. PMID:20696053

Implementation of collisions on GPU architecture in the Vorpal code

NASA Astrophysics Data System (ADS)

Leddy, Jarrod; Averkin, Sergey; Cowan, Ben; Sides, Scott; Werner, Greg; Cary, John

2017-10-01

The Vorpal code contains a variety of collision operators allowing for the simulation of plasmas containing multiple charge species interacting with neutrals, background gas, and EM fields. These existing algorithms have been improved and reimplemented to take advantage of the massive parallelization allowed by GPU architecture. The use of GPUs is most effective when algorithms are single-instruction multiple-data, so particle collisions are an ideal candidate for this parallelization technique due to their nature as a series of independent processes with the same underlying operation. This refactoring required data memory reorganization and careful consideration of device/host data allocation to minimize memory access and data communication per operation. Successful implementation has resulted in an order of magnitude increase in simulation speed for a test-case involving multiple binary collisions using the null collision method. Work supported by DARPA under contract W31P4Q-16-C-0009.

Improved look-up table method of computer-generated holograms.

PubMed

Wei, Hui; Gong, Guanghong; Li, Ni

2016-11-10

Heavy computation load and vast memory requirements are major bottlenecks of computer-generated holograms (CGHs), which are promising and challenging in three-dimensional displays. To solve these problems, an improved look-up table (LUT) method suitable for arbitrarily sampled object points is proposed and implemented on a graphics processing unit (GPU) whose reconstructed object quality is consistent with that of the coherent ray-trace (CRT) method. The concept of distance factor is defined, and the distance factors are pre-computed off-line and stored in a look-up table. The results show that while reconstruction quality close to that of the CRT method is obtained, the on-line computation time is dramatically reduced compared with the LUT method on the GPU and the memory usage is lower than that of the novel-LUT considerably. Optical experiments are carried out to validate the effectiveness of the proposed method.

gpuSPHASE-A shared memory caching implementation for 2D SPH using CUDA

NASA Astrophysics Data System (ADS)

Winkler, Daniel; Meister, Michael; Rezavand, Massoud; Rauch, Wolfgang

2017-04-01

Smoothed particle hydrodynamics (SPH) is a meshless Lagrangian method that has been successfully applied to computational fluid dynamics (CFD), solid mechanics and many other multi-physics problems. Using the method to solve transport phenomena in process engineering requires the simulation of several days to weeks of physical time. Based on the high computational demand of CFD such simulations in 3D need a computation time of years so that a reduction to a 2D domain is inevitable. In this paper gpuSPHASE, a new open-source 2D SPH solver implementation for graphics devices, is developed. It is optimized for simulations that must be executed with thousands of frames per second to be computed in reasonable time. A novel caching algorithm for Compute Unified Device Architecture (CUDA) shared memory is proposed and implemented. The software is validated and the performance is evaluated for the well established dambreak test case.

Real-time global illumination on mobile device

NASA Astrophysics Data System (ADS)

Ahn, Minsu; Ha, Inwoo; Lee, Hyong-Euk; Kim, James D. K.

2014-02-01

We propose a novel method for real-time global illumination on mobile devices. Our approach is based on instant radiosity, which uses a sequence of virtual point lights in order to represent the e ect of indirect illumination. Our rendering process consists of three stages. With the primary light, the rst stage generates a local illumination with the shadow map on GPU The second stage of the global illumination uses the re ective shadow map on GPU and generates the sequence of virtual point lights on CPU. Finally, we use the splatting method of Dachsbacher et al 1 and add the indirect illumination to the local illumination on GPU. With the limited computing resources in mobile devices, a small number of virtual point lights are allowed for real-time rendering. Our approach uses the multi-resolution sampling method with 3D geometry and attributes simultaneously and reduce the total number of virtual point lights. We also use the hybrid strategy, which collaboratively combines the CPUs and GPUs available in a mobile SoC due to the limited computing resources in mobile devices. Experimental results demonstrate the global illumination performance of the proposed method.

k(+)-buffer: An Efficient, Memory-Friendly and Dynamic k-buffer Framework.

PubMed

Vasilakis, Andreas-Alexandros; Papaioannou, Georgios; Fudos, Ioannis

2015-06-01

Depth-sorted fragment determination is fundamental for a host of image-based techniques which simulates complex rendering effects. It is also a challenging task in terms of time and space required when rasterizing scenes with high depth complexity. When low graphics memory requirements are of utmost importance, k-buffer can objectively be considered as the most preferred framework which advantageously ensures the correct depth order on a subset of all generated fragments. Although various alternatives have been introduced to partially or completely alleviate the noticeable quality artifacts produced by the initial k-buffer algorithm in the expense of memory increase or performance downgrade, appropriate tools to automatically and dynamically compute the most suitable value of k are still missing. To this end, we introduce k(+)-buffer, a fast framework that accurately simulates the behavior of k-buffer in a single rendering pass. Two memory-bounded data structures: (i) the max-array and (ii) the max-heap are developed on the GPU to concurrently maintain the k-foremost fragments per pixel by exploring pixel synchronization and fragment culling. Memory-friendly strategies are further introduced to dynamically (a) lessen the wasteful memory allocation of individual pixels with low depth complexity frequencies, (b) minimize the allocated size of k-buffer according to different application goals and hardware limitations via a straightforward depth histogram analysis and (c) manage local GPU cache with a fixed-memory depth-sorting mechanism. Finally, an extensive experimental evaluation is provided demonstrating the advantages of our work over all prior k-buffer variants in terms of memory usage, performance cost and image quality.

cuBLASTP: Fine-Grained Parallelization of Protein Sequence Search on CPU+GPU.

PubMed

Zhang, Jing; Wang, Hao; Feng, Wu-Chun

2017-01-01

BLAST, short for Basic Local Alignment Search Tool, is a ubiquitous tool used in the life sciences for pairwise sequence search. However, with the advent of next-generation sequencing (NGS), whether at the outset or downstream from NGS, the exponential growth of sequence databases is outstripping our ability to analyze the data. While recent studies have utilized the graphics processing unit (GPU) to speedup the BLAST algorithm for searching protein sequences (i.e., BLASTP), these studies use coarse-grained parallelism, where one sequence alignment is mapped to only one thread. Such an approach does not efficiently utilize the capabilities of a GPU, particularly due to the irregularity of BLASTP in both execution paths and memory-access patterns. To address the above shortcomings, we present a fine-grained approach to parallelize BLASTP, where each individual phase of sequence search is mapped to many threads on a GPU. This approach, which we refer to as cuBLASTP, reorders data-access patterns and reduces divergent branches of the most time-consuming phases (i.e., hit detection and ungapped extension). In addition, cuBLASTP optimizes the remaining phases (i.e., gapped extension and alignment with trace back) on a multicore CPU and overlaps their execution with the phases running on the GPU.

Parallelized multi–graphics processing unit framework for high-speed Gabor-domain optical coherence microscopy

PubMed Central

Tankam, Patrice; Santhanam, Anand P.; Lee, Kye-Sung; Won, Jungeun; Canavesi, Cristina; Rolland, Jannick P.

2014-01-01

Abstract. Gabor-domain optical coherence microscopy (GD-OCM) is a volumetric high-resolution technique capable of acquiring three-dimensional (3-D) skin images with histological resolution. Real-time image processing is needed to enable GD-OCM imaging in a clinical setting. We present a parallelized and scalable multi-graphics processing unit (GPU) computing framework for real-time GD-OCM image processing. A parallelized control mechanism was developed to individually assign computation tasks to each of the GPUs. For each GPU, the optimal number of amplitude-scans (A-scans) to be processed in parallel was selected to maximize GPU memory usage and core throughput. We investigated five computing architectures for computational speed-up in processing 1000×1000 A-scans. The proposed parallelized multi-GPU computing framework enables processing at a computational speed faster than the GD-OCM image acquisition, thereby facilitating high-speed GD-OCM imaging in a clinical setting. Using two parallelized GPUs, the image processing of a 1×1×0.6  mm3 skin sample was performed in about 13 s, and the performance was benchmarked at 6.5 s with four GPUs. This work thus demonstrates that 3-D GD-OCM data may be displayed in real-time to the examiner using parallelized GPU processing. PMID:24695868

Parallelized multi-graphics processing unit framework for high-speed Gabor-domain optical coherence microscopy.

PubMed

Tankam, Patrice; Santhanam, Anand P; Lee, Kye-Sung; Won, Jungeun; Canavesi, Cristina; Rolland, Jannick P

2014-07-01

Gabor-domain optical coherence microscopy (GD-OCM) is a volumetric high-resolution technique capable of acquiring three-dimensional (3-D) skin images with histological resolution. Real-time image processing is needed to enable GD-OCM imaging in a clinical setting. We present a parallelized and scalable multi-graphics processing unit (GPU) computing framework for real-time GD-OCM image processing. A parallelized control mechanism was developed to individually assign computation tasks to each of the GPUs. For each GPU, the optimal number of amplitude-scans (A-scans) to be processed in parallel was selected to maximize GPU memory usage and core throughput. We investigated five computing architectures for computational speed-up in processing 1000×1000 A-scans. The proposed parallelized multi-GPU computing framework enables processing at a computational speed faster than the GD-OCM image acquisition, thereby facilitating high-speed GD-OCM imaging in a clinical setting. Using two parallelized GPUs, the image processing of a 1×1×0.6  mm3 skin sample was performed in about 13 s, and the performance was benchmarked at 6.5 s with four GPUs. This work thus demonstrates that 3-D GD-OCM data may be displayed in real-time to the examiner using parallelized GPU processing.

GPU-accelerated low-latency real-time searches for gravitational waves from compact binary coalescence

NASA Astrophysics Data System (ADS)

Liu, Yuan; Du, Zhihui; Chung, Shin Kee; Hooper, Shaun; Blair, David; Wen, Linqing

2012-12-01

We present a graphics processing unit (GPU)-accelerated time-domain low-latency algorithm to search for gravitational waves (GWs) from coalescing binaries of compact objects based on the summed parallel infinite impulse response (SPIIR) filtering technique. The aim is to facilitate fast detection of GWs with a minimum delay to allow prompt electromagnetic follow-up observations. To maximize the GPU acceleration, we apply an efficient batched parallel computing model that significantly reduces the number of synchronizations in SPIIR and optimizes the usage of the memory and hardware resource. Our code is tested on the CUDA ‘Fermi’ architecture in a GTX 480 graphics card and its performance is compared with a single core of Intel Core i7 920 (2.67 GHz). A 58-fold speedup is achieved while giving results in close agreement with the CPU implementation. Our result indicates that it is possible to conduct a full search for GWs from compact binary coalescence in real time with only one desktop computer equipped with a Fermi GPU card for the initial LIGO detectors which in the past required more than 100 CPUs.

Architecting the Finite Element Method Pipeline for the GPU.

PubMed

Fu, Zhisong; Lewis, T James; Kirby, Robert M; Whitaker, Ross T

2014-02-01

The finite element method (FEM) is a widely employed numerical technique for approximating the solution of partial differential equations (PDEs) in various science and engineering applications. Many of these applications benefit from fast execution of the FEM pipeline. One way to accelerate the FEM pipeline is by exploiting advances in modern computational hardware, such as the many-core streaming processors like the graphical processing unit (GPU). In this paper, we present the algorithms and data-structures necessary to move the entire FEM pipeline to the GPU. First we propose an efficient GPU-based algorithm to generate local element information and to assemble the global linear system associated with the FEM discretization of an elliptic PDE. To solve the corresponding linear system efficiently on the GPU, we implement a conjugate gradient method preconditioned with a geometry-informed algebraic multi-grid (AMG) method preconditioner. We propose a new fine-grained parallelism strategy, a corresponding multigrid cycling stage and efficient data mapping to the many-core architecture of GPU. Comparison of our on-GPU assembly versus a traditional serial implementation on the CPU achieves up to an 87 × speedup. Focusing on the linear system solver alone, we achieve a speedup of up to 51 × versus use of a comparable state-of-the-art serial CPU linear system solver. Furthermore, the method compares favorably with other GPU-based, sparse, linear solvers.

Ray Casting of Large Multi-Resolution Volume Datasets

NASA Astrophysics Data System (ADS)

Lux, C.; Fröhlich, B.

2009-04-01

High quality volume visualization through ray casting on graphics processing units (GPU) has become an important approach for many application domains. We present a GPU-based, multi-resolution ray casting technique for the interactive visualization of massive volume data sets commonly found in the oil and gas industry. Large volume data sets are represented as a multi-resolution hierarchy based on an octree data structure. The original volume data is decomposed into small bricks of a fixed size acting as the leaf nodes of the octree. These nodes are the highest resolution of the volume. Coarser resolutions are represented through inner nodes of the hierarchy which are generated by down sampling eight neighboring nodes on a finer level. Due to limited memory resources of current desktop workstations and graphics hardware only a limited working set of bricks can be locally maintained for a frame to be displayed. This working set is chosen to represent the whole volume at different local resolution levels depending on the current viewer position, transfer function and distinct areas of interest. During runtime the working set of bricks is maintained in CPU- and GPU memory and is adaptively updated by asynchronously fetching data from external sources like hard drives or a network. The CPU memory hereby acts as a secondary level cache for these sources from which the GPU representation is updated. Our volume ray casting algorithm is based on a 3D texture-atlas in GPU memory. This texture-atlas contains the complete working set of bricks of the current multi-resolution representation of the volume. This enables the volume ray casting algorithm to access the whole working set of bricks through only a single 3D texture. For traversing rays through the volume, information about the locations and resolution levels of visited bricks are required for correct compositing computations. We encode this information into a small 3D index texture which represents the current octree subdivision on its finest level and spatially organizes the bricked data. This approach allows us to render a bricked multi-resolution volume data set utilizing only a single rendering pass with no loss of compositing precision. In contrast most state-of-the art volume rendering systems handle the bricked data as individual 3D textures, which are rendered one at a time while the results are composited into a lower precision frame buffer. Furthermore, our method enables us to integrate advanced volume rendering techniques like empty-space skipping, adaptive sampling and preintegrated transfer functions in a very straightforward manner with virtually no extra costs. Our interactive volume ray tracing implementation allows high quality visualizations of massive volume data sets of tens of Gigabytes in size on standard desktop workstations.

Accelerate quasi Monte Carlo method for solving systems of linear algebraic equations through shared memory

NASA Astrophysics Data System (ADS)

Lai, Siyan; Xu, Ying; Shao, Bo; Guo, Menghan; Lin, Xiaola

2017-04-01

In this paper we study on Monte Carlo method for solving systems of linear algebraic equations (SLAE) based on shared memory. Former research demostrated that GPU can effectively speed up the computations of this issue. Our purpose is to optimize Monte Carlo method simulation on GPUmemoryachritecture specifically. Random numbers are organized to storein shared memory, which aims to accelerate the parallel algorithm. Bank conflicts can be avoided by our Collaborative Thread Arrays(CTA)scheme. The results of experiments show that the shared memory based strategy can speed up the computaions over than 3X at most.

Power/Performance Trade-offs of Small Batched LU Based Solvers on GPUs

DOE Office of Scientific and Technical Information (OSTI.GOV)

Villa, Oreste; Fatica, Massimiliano; Gawande, Nitin A.

In this paper we propose and analyze a set of batched linear solvers for small matrices on Graphic Processing Units (GPUs), evaluating the various alternatives depending on the size of the systems to solve. We discuss three different solutions that operate with different level of parallelization and GPU features. The first, exploiting the CUBLAS library, manages matrices of size up to 32x32 and employs Warp level (one matrix, one Warp) parallelism and shared memory. The second works at Thread-block level parallelism (one matrix, one Thread-block), still exploiting shared memory but managing matrices up to 76x76. The third is Thread levelmore » parallel (one matrix, one thread) and can reach sizes up to 128x128, but it does not exploit shared memory and only relies on the high memory bandwidth of the GPU. The first and second solution only support partial pivoting, the third one easily supports partial and full pivoting, making it attractive to problems that require greater numerical stability. We analyze the trade-offs in terms of performance and power consumption as function of the size of the linear systems that are simultaneously solved. We execute the three implementations on a Tesla M2090 (Fermi) and on a Tesla K20 (Kepler).« less

CUDA Fortran acceleration for the finite-difference time-domain method

NASA Astrophysics Data System (ADS)

Hadi, Mohammed F.; Esmaeili, Seyed A.

2013-05-01

A detailed description of programming the three-dimensional finite-difference time-domain (FDTD) method to run on graphical processing units (GPUs) using CUDA Fortran is presented. Two FDTD-to-CUDA thread-block mapping designs are investigated and their performances compared. Comparative assessment of trade-offs between GPU's shared memory and L1 cache is also discussed. This presentation is for the benefit of FDTD programmers who work exclusively with Fortran and are reluctant to port their codes to C in order to utilize GPU computing. The derived CUDA Fortran code is compared with an optimized CPU version that runs on a workstation-class CPU to present a realistic GPU to CPU run time comparison and thus help in making better informed investment decisions on FDTD code redesigns and equipment upgrades. All analyses are mirrored with CUDA C simulations to put in perspective the present state of CUDA Fortran development.

Montblanc1: GPU accelerated radio interferometer measurement equations in support of Bayesian inference for radio observations

NASA Astrophysics Data System (ADS)

Perkins, S. J.; Marais, P. C.; Zwart, J. T. L.; Natarajan, I.; Tasse, C.; Smirnov, O.

2015-09-01

We present Montblanc, a GPU implementation of the Radio interferometer measurement equation (RIME) in support of the Bayesian inference for radio observations (BIRO) technique. BIRO uses Bayesian inference to select sky models that best match the visibilities observed by a radio interferometer. To accomplish this, BIRO evaluates the RIME multiple times, varying sky model parameters to produce multiple model visibilities. χ2 values computed from the model and observed visibilities are used as likelihood values to drive the Bayesian sampling process and select the best sky model. As most of the elements of the RIME and χ2 calculation are independent of one another, they are highly amenable to parallel computation. Additionally, Montblanc caters for iterative RIME evaluation to produce multiple χ2 values. Modified model parameters are transferred to the GPU between each iteration. We implemented Montblanc as a Python package based upon NVIDIA's CUDA architecture. As such, it is easy to extend and implement different pipelines. At present, Montblanc supports point and Gaussian morphologies, but is designed for easy addition of new source profiles. Montblanc's RIME implementation is performant: On an NVIDIA K40, it is approximately 250 times faster than MEQTREES on a dual hexacore Intel E5-2620v2 CPU. Compared to the OSKAR simulator's GPU-implemented RIME components it is 7.7 and 12 times faster on the same K40 for single and double-precision floating point respectively. However, OSKAR's RIME implementation is more general than Montblanc's BIRO-tailored RIME. Theoretical analysis of Montblanc's dominant CUDA kernel suggests that it is memory bound. In practice, profiling shows that is balanced between compute and memory, as much of the data required by the problem is retained in L1 and L2 caches.

SU-E-T-500: Initial Implementation of GPU-Based Particle Swarm Optimization for 4D IMRT Planning in Lung SBRT

DOE Office of Scientific and Technical Information (OSTI.GOV)

Modiri, A; Hagan, A; Gu, X

Purpose 4D-IMRT planning, combined with dynamic MLC tracking delivery, utilizes the temporal dimension as an additional degree of freedom to achieve improved OAR-sparing. The computational complexity for such optimization increases exponentially with increase in dimensionality. In order to accomplish this task in a clinically-feasible time frame, we present an initial implementation of GPU-based 4D-IMRT planning based on particle swarm optimization (PSO). Methods The target and normal structures were manually contoured on ten phases of a 4DCT scan of a NSCLC patient with a 54cm3 right-lower-lobe tumor (1.5cm motion). Corresponding ten 3D-IMRT plans were created in the Eclipse treatment planning systemmore » (Ver-13.6). A vendor-provided scripting interface was used to export 3D-dose matrices corresponding to each control point (10 phases × 9 beams × 166 control points = 14,940), which served as input to PSO. The optimization task was to iteratively adjust the weights of each control point and scale the corresponding dose matrices. In order to handle the large amount of data in GPU memory, dose matrices were sparsified and placed in contiguous memory blocks with the 14,940 weight-variables. PSO was implemented on CPU (dual-Xeon, 3.1GHz) and GPU (dual-K20 Tesla, 2496 cores, 3.52Tflops, each) platforms. NiftyReg, an open-source deformable image registration package, was used to calculate the summed dose. Results The 4D-PSO plan yielded PTV coverage comparable to the clinical ITV-based plan and significantly higher OAR-sparing, as follows: lung Dmean=33%; lung V20=27%; spinal cord Dmax=26%; esophagus Dmax=42%; heart Dmax=0%; heart Dmean=47%. The GPU-PSO processing time for 14940 variables and 7 PSO-particles was 41% that of CPU-PSO (199 vs. 488 minutes). Conclusion Truly 4D-IMRT planning can yield significant OAR dose-sparing while preserving PTV coverage. The corresponding optimization problem is large-scale, non-convex and computationally rigorous. Our initial results indicate that GPU-based PSO with further software optimization can make such planning clinically feasible. This work was supported through funding from the National Institutes of Health and Varian Medical Systems.« less

Graphics Processing Units (GPU) and the Goddard Earth Observing System atmospheric model (GEOS-5): Implementation and Potential Applications

NASA Technical Reports Server (NTRS)

Putnam, William M.

2011-01-01

Earth system models like the Goddard Earth Observing System model (GEOS-5) have been pushing the limits of large clusters of multi-core microprocessors, producing breath-taking fidelity in resolving cloud systems at a global scale. GPU computing presents an opportunity for improving the efficiency of these leading edge models. A GPU implementation of GEOS-5 will facilitate the use of cloud-system resolving resolutions in data assimilation and weather prediction, at resolutions near 3.5 km, improving our ability to extract detailed information from high-resolution satellite observations and ultimately produce better weather and climate predictions

GPU Linear Algebra Libraries and GPGPU Programming for Accelerating MOPAC Semiempirical Quantum Chemistry Calculations.

PubMed

Maia, Julio Daniel Carvalho; Urquiza Carvalho, Gabriel Aires; Mangueira, Carlos Peixoto; Santana, Sidney Ramos; Cabral, Lucidio Anjos Formiga; Rocha, Gerd B

2012-09-11

In this study, we present some modifications in the semiempirical quantum chemistry MOPAC2009 code that accelerate single-point energy calculations (1SCF) of medium-size (up to 2500 atoms) molecular systems using GPU coprocessors and multithreaded shared-memory CPUs. Our modifications consisted of using a combination of highly optimized linear algebra libraries for both CPU (LAPACK and BLAS from Intel MKL) and GPU (MAGMA and CUBLAS) to hasten time-consuming parts of MOPAC such as the pseudodiagonalization, full diagonalization, and density matrix assembling. We have shown that it is possible to obtain large speedups just by using CPU serial linear algebra libraries in the MOPAC code. As a special case, we show a speedup of up to 14 times for a methanol simulation box containing 2400 atoms and 4800 basis functions, with even greater gains in performance when using multithreaded CPUs (2.1 times in relation to the single-threaded CPU code using linear algebra libraries) and GPUs (3.8 times). This degree of acceleration opens new perspectives for modeling larger structures which appear in inorganic chemistry (such as zeolites and MOFs), biochemistry (such as polysaccharides, small proteins, and DNA fragments), and materials science (such as nanotubes and fullerenes). In addition, we believe that this parallel (GPU-GPU) MOPAC code will make it feasible to use semiempirical methods in lengthy molecular simulations using both hybrid QM/MM and QM/QM potentials.

High-throughput sequence alignment using Graphics Processing Units

PubMed Central

Schatz, Michael C; Trapnell, Cole; Delcher, Arthur L; Varshney, Amitabh

2007-01-01

Background The recent availability of new, less expensive high-throughput DNA sequencing technologies has yielded a dramatic increase in the volume of sequence data that must be analyzed. These data are being generated for several purposes, including genotyping, genome resequencing, metagenomics, and de novo genome assembly projects. Sequence alignment programs such as MUMmer have proven essential for analysis of these data, but researchers will need ever faster, high-throughput alignment tools running on inexpensive hardware to keep up with new sequence technologies. Results This paper describes MUMmerGPU, an open-source high-throughput parallel pairwise local sequence alignment program that runs on commodity Graphics Processing Units (GPUs) in common workstations. MUMmerGPU uses the new Compute Unified Device Architecture (CUDA) from nVidia to align multiple query sequences against a single reference sequence stored as a suffix tree. By processing the queries in parallel on the highly parallel graphics card, MUMmerGPU achieves more than a 10-fold speedup over a serial CPU version of the sequence alignment kernel, and outperforms the exact alignment component of MUMmer on a high end CPU by 3.5-fold in total application time when aligning reads from recent sequencing projects using Solexa/Illumina, 454, and Sanger sequencing technologies. Conclusion MUMmerGPU is a low cost, ultra-fast sequence alignment program designed to handle the increasing volume of data produced by new, high-throughput sequencing technologies. MUMmerGPU demonstrates that even memory-intensive applications can run significantly faster on the relatively low-cost GPU than on the CPU. PMID:18070356

Analysis and Implementation of Particle-to-Particle (P2P) Graphics Processor Unit (GPU) Kernel for Black-Box Adaptive Fast Multipole Method

DTIC Science & Technology

2015-06-01

5110P and 16 dx360M4 nodes each with one NVIDIA Kepler K20M/K40M GPU. Each node contained dual Intel Xeon E5-2670 (Sandy Bridge) central processing...kernel and as such does not employ multiple processors. This work makes use of a single processing core and a single NVIDIA Kepler K40 GK110...bandwidth (2 × 16 slot), 7.877 GFloat/s; Kepler K40 peak, 4,290 × 1 billion floating-point operations (GFLOPs), and 288 GB/s Kepler K40 memory

Molecular dynamics simulations through GPU video games technologies

PubMed Central

Loukatou, Styliani; Papageorgiou, Louis; Fakourelis, Paraskevas; Filntisi, Arianna; Polychronidou, Eleftheria; Bassis, Ioannis; Megalooikonomou, Vasileios; Makałowski, Wojciech; Vlachakis, Dimitrios; Kossida, Sophia

2016-01-01

Bioinformatics is the scientific field that focuses on the application of computer technology to the management of biological information. Over the years, bioinformatics applications have been used to store, process and integrate biological and genetic information, using a wide range of methodologies. One of the most de novo techniques used to understand the physical movements of atoms and molecules is molecular dynamics (MD). MD is an in silico method to simulate the physical motions of atoms and molecules under certain conditions. This has become a state strategic technique and now plays a key role in many areas of exact sciences, such as chemistry, biology, physics and medicine. Due to their complexity, MD calculations could require enormous amounts of computer memory and time and therefore their execution has been a big problem. Despite the huge computational cost, molecular dynamics have been implemented using traditional computers with a central memory unit (CPU). A graphics processing unit (GPU) computing technology was first designed with the goal to improve video games, by rapidly creating and displaying images in a frame buffer such as screens. The hybrid GPU-CPU implementation, combined with parallel computing is a novel technology to perform a wide range of calculations. GPUs have been proposed and used to accelerate many scientific computations including MD simulations. Herein, we describe the new methodologies developed initially as video games and how they are now applied in MD simulations. PMID:27525251

Design and Implementation of High-Performance GIS Dynamic Objects Rendering Engine

NASA Astrophysics Data System (ADS)

Zhong, Y.; Wang, S.; Li, R.; Yun, W.; Song, G.

2017-12-01

Spatio-temporal dynamic visualization is more vivid than static visualization. It important to use dynamic visualization techniques to reveal the variation process and trend vividly and comprehensively for the geographical phenomenon. To deal with challenges caused by dynamic visualization of both 2D and 3D spatial dynamic targets, especially for different spatial data types require high-performance GIS dynamic objects rendering engine. The main approach for improving the rendering engine with vast dynamic targets relies on key technologies of high-performance GIS, including memory computing, parallel computing, GPU computing and high-performance algorisms. In this study, high-performance GIS dynamic objects rendering engine is designed and implemented for solving the problem based on hybrid accelerative techniques. The high-performance GIS rendering engine contains GPU computing, OpenGL technology, and high-performance algorism with the advantage of 64-bit memory computing. It processes 2D, 3D dynamic target data efficiently and runs smoothly with vast dynamic target data. The prototype system of high-performance GIS dynamic objects rendering engine is developed based SuperMap GIS iObjects. The experiments are designed for large-scale spatial data visualization, the results showed that the high-performance GIS dynamic objects rendering engine have the advantage of high performance. Rendering two-dimensional and three-dimensional dynamic objects achieve 20 times faster on GPU than on CPU.

Accelerating solutions of one-dimensional unsteady PDEs with GPU-based swept time-space decomposition

NASA Astrophysics Data System (ADS)

Magee, Daniel J.; Niemeyer, Kyle E.

2018-03-01

The expedient design of precision components in aerospace and other high-tech industries requires simulations of physical phenomena often described by partial differential equations (PDEs) without exact solutions. Modern design problems require simulations with a level of resolution difficult to achieve in reasonable amounts of time-even in effectively parallelized solvers. Though the scale of the problem relative to available computing power is the greatest impediment to accelerating these applications, significant performance gains can be achieved through careful attention to the details of memory communication and access. The swept time-space decomposition rule reduces communication between sub-domains by exhausting the domain of influence before communicating boundary values. Here we present a GPU implementation of the swept rule, which modifies the algorithm for improved performance on this processing architecture by prioritizing use of private (shared) memory, avoiding interblock communication, and overwriting unnecessary values. It shows significant improvement in the execution time of finite-difference solvers for one-dimensional unsteady PDEs, producing speedups of 2 - 9 × for a range of problem sizes, respectively, compared with simple GPU versions and 7 - 300 × compared with parallel CPU versions. However, for a more sophisticated one-dimensional system of equations discretized with a second-order finite-volume scheme, the swept rule performs 1.2 - 1.9 × worse than a standard implementation for all problem sizes.

Multilevel Summation of Electrostatic Potentials Using Graphics Processing Units*

PubMed Central

Hardy, David J.; Stone, John E.; Schulten, Klaus

2009-01-01

Physical and engineering practicalities involved in microprocessor design have resulted in flat performance growth for traditional single-core microprocessors. The urgent need for continuing increases in the performance of scientific applications requires the use of many-core processors and accelerators such as graphics processing units (GPUs). This paper discusses GPU acceleration of the multilevel summation method for computing electrostatic potentials and forces for a system of charged atoms, which is a problem of paramount importance in biomolecular modeling applications. We present and test a new GPU algorithm for the long-range part of the potentials that computes a cutoff pair potential between lattice points, essentially convolving a fixed 3-D lattice of “weights” over all sub-cubes of a much larger lattice. The implementation exploits the different memory subsystems provided on the GPU to stream optimally sized data sets through the multiprocessors. We demonstrate for the full multilevel summation calculation speedups of up to 26 using a single GPU and 46 using multiple GPUs, enabling the computation of a high-resolution map of the electrostatic potential for a system of 1.5 million atoms in under 12 seconds. PMID:20161132

Matrix decomposition graphics processing unit solver for Poisson image editing

NASA Astrophysics Data System (ADS)

Lei, Zhao; Wei, Li

2012-10-01

In recent years, gradient-domain methods have been widely discussed in the image processing field, including seamless cloning and image stitching. These algorithms are commonly carried out by solving a large sparse linear system: the Poisson equation. However, solving the Poisson equation is a computational and memory intensive task which makes it not suitable for real-time image editing. A new matrix decomposition graphics processing unit (GPU) solver (MDGS) is proposed to settle the problem. A matrix decomposition method is used to distribute the work among GPU threads, so that MDGS will take full advantage of the computing power of current GPUs. Additionally, MDGS is a hybrid solver (combines both the direct and iterative techniques) and has two-level architecture. These enable MDGS to generate identical solutions with those of the common Poisson methods and achieve high convergence rate in most cases. This approach is advantageous in terms of parallelizability, enabling real-time image processing, low memory-taken and extensive applications.

GPU acceleration of particle-in-cell methods

NASA Astrophysics Data System (ADS)

Cowan, Benjamin; Cary, John; Meiser, Dominic

2015-11-01

Graphics processing units (GPUs) have become key components in many supercomputing systems, as they can provide more computations relative to their cost and power consumption than conventional processors. However, to take full advantage of this capability, they require a strict programming model which involves single-instruction multiple-data execution as well as significant constraints on memory accesses. To bring the full power of GPUs to bear on plasma physics problems, we must adapt the computational methods to this new programming model. We have developed a GPU implementation of the particle-in-cell (PIC) method, one of the mainstays of plasma physics simulation. This framework is highly general and enables advanced PIC features such as high order particles and absorbing boundary conditions. The main elements of the PIC loop, including field interpolation and particle deposition, are designed to optimize memory access. We describe the performance of these algorithms and discuss some of the methods used. Work supported by DARPA contract W31P4Q-15-C-0061 (SBIR).

Real-time autocorrelator for fluorescence correlation spectroscopy based on graphical-processor-unit architecture: method, implementation, and comparative studies

NASA Astrophysics Data System (ADS)

Laracuente, Nicholas; Grossman, Carl

2013-03-01

We developed an algorithm and software to calculate autocorrelation functions from real-time photon-counting data using the fast, parallel capabilities of graphical processor units (GPUs). Recent developments in hardware and software have allowed for general purpose computing with inexpensive GPU hardware. These devices are more suited for emulating hardware autocorrelators than traditional CPU-based software applications by emphasizing parallel throughput over sequential speed. Incoming data are binned in a standard multi-tau scheme with configurable points-per-bin size and are mapped into a GPU memory pattern to reduce time-expensive memory access. Applications include dynamic light scattering (DLS) and fluorescence correlation spectroscopy (FCS) experiments. We ran the software on a 64-core graphics pci card in a 3.2 GHz Intel i5 CPU based computer running Linux. FCS measurements were made on Alexa-546 and Texas Red dyes in a standard buffer (PBS). Software correlations were compared to hardware correlator measurements on the same signals. Supported by HHMI and Swarthmore College

Extending the BEAGLE library to a multi-FPGA platform.

PubMed

Jin, Zheming; Bakos, Jason D

2013-01-19

Maximum Likelihood (ML)-based phylogenetic inference using Felsenstein's pruning algorithm is a standard method for estimating the evolutionary relationships amongst a set of species based on DNA sequence data, and is used in popular applications such as RAxML, PHYLIP, GARLI, BEAST, and MrBayes. The Phylogenetic Likelihood Function (PLF) and its associated scaling and normalization steps comprise the computational kernel for these tools. These computations are data intensive but contain fine grain parallelism that can be exploited by coprocessor architectures such as FPGAs and GPUs. A general purpose API called BEAGLE has recently been developed that includes optimized implementations of Felsenstein's pruning algorithm for various data parallel architectures. In this paper, we extend the BEAGLE API to a multiple Field Programmable Gate Array (FPGA)-based platform called the Convey HC-1. The core calculation of our implementation, which includes both the phylogenetic likelihood function (PLF) and the tree likelihood calculation, has an arithmetic intensity of 130 floating-point operations per 64 bytes of I/O, or 2.03 ops/byte. Its performance can thus be calculated as a function of the host platform's peak memory bandwidth and the implementation's memory efficiency, as 2.03 × peak bandwidth × memory efficiency. Our FPGA-based platform has a peak bandwidth of 76.8 GB/s and our implementation achieves a memory efficiency of approximately 50%, which gives an average throughput of 78 Gflops. This represents a ~40X speedup when compared with BEAGLE's CPU implementation on a dual Xeon 5520 and 3X speedup versus BEAGLE's GPU implementation on a Tesla T10 GPU for very large data sizes. The power consumption is 92 W, yielding a power efficiency of 1.7 Gflops per Watt. The use of data parallel architectures to achieve high performance for likelihood-based phylogenetic inference requires high memory bandwidth and a design methodology that emphasizes high memory efficiency. To achieve this objective, we integrated 32 pipelined processing elements (PEs) across four FPGAs. For the design of each PE, we developed a specialized synthesis tool to generate a floating-point pipeline with resource and throughput constraints to match the target platform. We have found that using low-latency floating-point operators can significantly reduce FPGA area and still meet timing requirement on the target platform. We found that this design methodology can achieve performance that exceeds that of a GPU-based coprocessor.

Computing the Density Matrix in Electronic Structure Theory on Graphics Processing Units.

PubMed

Cawkwell, M J; Sanville, E J; Mniszewski, S M; Niklasson, Anders M N

2012-11-13

The self-consistent solution of a Schrödinger-like equation for the density matrix is a critical and computationally demanding step in quantum-based models of interatomic bonding. This step was tackled historically via the diagonalization of the Hamiltonian. We have investigated the performance and accuracy of the second-order spectral projection (SP2) algorithm for the computation of the density matrix via a recursive expansion of the Fermi operator in a series of generalized matrix-matrix multiplications. We demonstrate that owing to its simplicity, the SP2 algorithm [Niklasson, A. M. N. Phys. Rev. B2002, 66, 155115] is exceptionally well suited to implementation on graphics processing units (GPUs). The performance in double and single precision arithmetic of a hybrid GPU/central processing unit (CPU) and full GPU implementation of the SP2 algorithm exceed those of a CPU-only implementation of the SP2 algorithm and traditional matrix diagonalization when the dimensions of the matrices exceed about 2000 × 2000. Padding schemes for arrays allocated in the GPU memory that optimize the performance of the CUBLAS implementations of the level 3 BLAS DGEMM and SGEMM subroutines for generalized matrix-matrix multiplications are described in detail. The analysis of the relative performance of the hybrid CPU/GPU and full GPU implementations indicate that the transfer of arrays between the GPU and CPU constitutes only a small fraction of the total computation time. The errors measured in the self-consistent density matrices computed using the SP2 algorithm are generally smaller than those measured in matrices computed via diagonalization. Furthermore, the errors in the density matrices computed using the SP2 algorithm do not exhibit any dependence of system size, whereas the errors increase linearly with the number of orbitals when diagonalization is employed.

Fast Simulation of Dynamic Ultrasound Images Using the GPU.

PubMed

Storve, Sigurd; Torp, Hans

2017-10-01

Simulated ultrasound data is a valuable tool for development and validation of quantitative image analysis methods in echocardiography. Unfortunately, simulation time can become prohibitive for phantoms consisting of a large number of point scatterers. The COLE algorithm by Gao et al. is a fast convolution-based simulator that trades simulation accuracy for improved speed. We present highly efficient parallelized CPU and GPU implementations of the COLE algorithm with an emphasis on dynamic simulations involving moving point scatterers. We argue that it is crucial to minimize the amount of data transfers from the CPU to achieve good performance on the GPU. We achieve this by storing the complete trajectories of the dynamic point scatterers as spline curves in the GPU memory. This leads to good efficiency when simulating sequences consisting of a large number of frames, such as B-mode and tissue Doppler data for a full cardiac cycle. In addition, we propose a phase-based subsample delay technique that efficiently eliminates flickering artifacts seen in B-mode sequences when COLE is used without enough temporal oversampling. To assess the performance, we used a laptop computer and a desktop computer, each equipped with a multicore Intel CPU and an NVIDIA GPU. Running the simulator on a high-end TITAN X GPU, we observed two orders of magnitude speedup compared to the parallel CPU version, three orders of magnitude speedup compared to simulation times reported by Gao et al. in their paper on COLE, and a speedup of 27000 times compared to the multithreaded version of Field II, using numbers reported in a paper by Jensen. We hope that by releasing the simulator as an open-source project we will encourage its use and further development.

Scaling deep learning on GPU and knights landing clusters

DOE Office of Scientific and Technical Information (OSTI.GOV)

You, Yang; Buluc, Aydin; Demmel, James

Training neural networks has become a big bottleneck. For example, training ImageNet dataset on one Nvidia K20 GPU needs 21 days. To speed up the training process, the current deep learning systems heavily rely on the hardware accelerators. However, these accelerators have limited on-chip memory compared with CPUs. We use both self-host Intel Knights Landing (KNL) clusters and multi-GPU clusters as our target platforms. From the algorithm aspect, we focus on Elastic Averaging SGD (EASGD) to design algorithms for HPC clusters. We redesign four efficient algorithms for HPC systems to improve EASGD's poor scaling on clusters. Async EASGD, Async MEASGD,more » and Hogwild EASGD are faster than existing counter-part methods (Async SGD, Async MSGD, and Hogwild SGD) in all comparisons. Sync EASGD achieves 5.3X speedup over original EASGD on the same platform. We achieve 91.5% weak scaling efficiency on 4253 KNL cores, which is higher than the state-of-the-art implementation.« less

NaNet: a configurable NIC bridging the gap between HPC and real-time HEP GPU computing

NASA Astrophysics Data System (ADS)

Lonardo, A.; Ameli, F.; Ammendola, R.; Biagioni, A.; Cotta Ramusino, A.; Fiorini, M.; Frezza, O.; Lamanna, G.; Lo Cicero, F.; Martinelli, M.; Neri, I.; Paolucci, P. S.; Pastorelli, E.; Pontisso, L.; Rossetti, D.; Simeone, F.; Simula, F.; Sozzi, M.; Tosoratto, L.; Vicini, P.

2015-04-01

NaNet is a FPGA-based PCIe Network Interface Card (NIC) design with GPUDirect and Remote Direct Memory Access (RDMA) capabilities featuring a configurable and extensible set of network channels. The design currently supports both standard—Gbe (1000BASE-T) and 10GbE (10Base-R)—and custom—34 Gbps APElink and 2.5 Gbps deterministic latency KM3link—channels, but its modularity allows for straightforward inclusion of other link technologies. The GPUDirect feature combined with a transport layer offload module and a data stream processing stage makes NaNet a low-latency NIC suitable for real-time GPU processing. In this paper we describe the NaNet architecture and its performances, exhibiting two of its use cases: the GPU-based low-level trigger for the RICH detector in the NA62 experiment at CERN and the on-/off-shore data transport system for the KM3NeT-IT underwater neutrino telescope.

On the Usage of GPUs for Efficient Motion Estimation in Medical Image Sequences

PubMed Central

Thiyagalingam, Jeyarajan; Goodman, Daniel; Schnabel, Julia A.; Trefethen, Anne; Grau, Vicente

2011-01-01

Images are ubiquitous in biomedical applications from basic research to clinical practice. With the rapid increase in resolution, dimensionality of the images and the need for real-time performance in many applications, computational requirements demand proper exploitation of multicore architectures. Towards this, GPU-specific implementations of image analysis algorithms are particularly promising. In this paper, we investigate the mapping of an enhanced motion estimation algorithm to novel GPU-specific architectures, the resulting challenges and benefits therein. Using a database of three-dimensional image sequences, we show that the mapping leads to substantial performance gains, up to a factor of 60, and can provide near-real-time experience. We also show how architectural peculiarities of these devices can be best exploited in the benefit of algorithms, most specifically for addressing the challenges related to their access patterns and different memory configurations. Finally, we evaluate the performance of the algorithm on three different GPU architectures and perform a comprehensive analysis of the results. PMID:21869880

A GPU accelerated PDF transparency engine

NASA Astrophysics Data System (ADS)

Recker, John; Lin, I.-Jong; Tastl, Ingeborg

2011-01-01

As commercial printing presses become faster, cheaper and more efficient, so too must the Raster Image Processors (RIP) that prepare data for them to print. Digital press RIPs, however, have been challenged to on the one hand meet the ever increasing print performance of the latest digital presses, and on the other hand process increasingly complex documents with transparent layers and embedded ICC profiles. This paper explores the challenges encountered when implementing a GPU accelerated driver for the open source Ghostscript Adobe PostScript and PDF language interpreter targeted at accelerating PDF transparency for high speed commercial presses. It further describes our solution, including an image memory manager for tiling input and output images and documents, a PDF compatible multiple image layer blending engine, and a GPU accelerated ICC v4 compatible color transformation engine. The result, we believe, is the foundation for a scalable, efficient, distributed RIP system that can meet current and future RIP requirements for a wide range of commercial digital presses.

Track finding in ATLAS using GPUs

NASA Astrophysics Data System (ADS)

Mattmann, J.; Schmitt, C.

2012-12-01

The reconstruction and simulation of collision events is a major task in modern HEP experiments involving several ten thousands of standard CPUs. On the other hand the graphics processors (GPUs) have become much more powerful and are by far outperforming the standard CPUs in terms of floating point operations due to their massive parallel approach. The usage of these GPUs could therefore significantly reduce the overall reconstruction time per event or allow for the usage of more sophisticated algorithms. In this paper the track finding in the ATLAS experiment will be used as an example on how the GPUs can be used in this context: the implementation on the GPU requires a change in the algorithmic flow to allow the code to work in the rather limited environment on the GPU in terms of memory, cache, and transfer speed from and to the GPU and to make use of the massive parallel computation. Both, the specific implementation of parts of the ATLAS track reconstruction chain and the performance improvements obtained will be discussed.

DOE Office of Scientific and Technical Information (OSTI.GOV)

Lindstrom, P; Cohen, J D

We present a streaming geometry compression codec for multiresolution, uniformly-gridded, triangular terrain patches that supports very fast decompression. Our method is based on linear prediction and residual coding for lossless compression of the full-resolution data. As simplified patches on coarser levels in the hierarchy already incur some data loss, we optionally allow further quantization for more lossy compression. The quantization levels are adaptive on a per-patch basis, while still permitting seamless, adaptive tessellations of the terrain. Our geometry compression on such a hierarchy achieves compression ratios of 3:1 to 12:1. Our scheme is not only suitable for fast decompression onmore » the CPU, but also for parallel decoding on the GPU with peak throughput over 2 billion triangles per second. Each terrain patch is independently decompressed on the fly from a variable-rate bitstream by a GPU geometry program with no branches or conditionals. Thus we can store the geometry compressed on the GPU, reducing storage and bandwidth requirements throughout the system. In our rendering approach, only compressed bitstreams and the decoded height values in the view-dependent 'cut' are explicitly stored on the GPU. Normal vectors are computed in a streaming fashion, and remaining geometry and texture coordinates, as well as mesh connectivity, are shared and re-used for all patches. We demonstrate and evaluate our algorithms on a small prototype system in which all compressed geometry fits in the GPU memory and decompression occurs on the fly every rendering frame without any cache maintenance.« less

GPU-computing in econophysics and statistical physics

NASA Astrophysics Data System (ADS)

Preis, T.

2011-03-01

A recent trend in computer science and related fields is general purpose computing on graphics processing units (GPUs), which can yield impressive performance. With multiple cores connected by high memory bandwidth, today's GPUs offer resources for non-graphics parallel processing. This article provides a brief introduction into the field of GPU computing and includes examples. In particular computationally expensive analyses employed in financial market context are coded on a graphics card architecture which leads to a significant reduction of computing time. In order to demonstrate the wide range of possible applications, a standard model in statistical physics - the Ising model - is ported to a graphics card architecture as well, resulting in large speedup values.

MILC Code Performance on High End CPU and GPU Supercomputer Clusters

NASA Astrophysics Data System (ADS)

DeTar, Carleton; Gottlieb, Steven; Li, Ruizi; Toussaint, Doug

2018-03-01

With recent developments in parallel supercomputing architecture, many core, multi-core, and GPU processors are now commonplace, resulting in more levels of parallelism, memory hierarchy, and programming complexity. It has been necessary to adapt the MILC code to these new processors starting with NVIDIA GPUs, and more recently, the Intel Xeon Phi processors. We report on our efforts to port and optimize our code for the Intel Knights Landing architecture. We consider performance of the MILC code with MPI and OpenMP, and optimizations with QOPQDP and QPhiX. For the latter approach, we concentrate on the staggered conjugate gradient and gauge force. We also consider performance on recent NVIDIA GPUs using the QUDA library.

GPU-accelerated algorithms for many-particle continuous-time quantum walks

NASA Astrophysics Data System (ADS)

Piccinini, Enrico; Benedetti, Claudia; Siloi, Ilaria; Paris, Matteo G. A.; Bordone, Paolo

2017-06-01

Many-particle continuous-time quantum walks (CTQWs) represent a resource for several tasks in quantum technology, including quantum search algorithms and universal quantum computation. In order to design and implement CTQWs in a realistic scenario, one needs effective simulation tools for Hamiltonians that take into account static noise and fluctuations in the lattice, i.e. Hamiltonians containing stochastic terms. To this aim, we suggest a parallel algorithm based on the Taylor series expansion of the evolution operator, and compare its performances with those of algorithms based on the exact diagonalization of the Hamiltonian or a 4th order Runge-Kutta integration. We prove that both Taylor-series expansion and Runge-Kutta algorithms are reliable and have a low computational cost, the Taylor-series expansion showing the additional advantage of a memory allocation not depending on the precision of calculation. Both algorithms are also highly parallelizable within the SIMT paradigm, and are thus suitable for GPGPU computing. In turn, we have benchmarked 4 NVIDIA GPUs and 3 quad-core Intel CPUs for a 2-particle system over lattices of increasing dimension, showing that the speedup provided by GPU computing, with respect to the OPENMP parallelization, lies in the range between 8x and (more than) 20x, depending on the frequency of post-processing. GPU-accelerated codes thus allow one to overcome concerns about the execution time, and make it possible simulations with many interacting particles on large lattices, with the only limit of the memory available on the device.

BALSA: integrated secondary analysis for whole-genome and whole-exome sequencing, accelerated by GPU.

PubMed

Luo, Ruibang; Wong, Yiu-Lun; Law, Wai-Chun; Lee, Lap-Kei; Cheung, Jeanno; Liu, Chi-Man; Lam, Tak-Wah

2014-01-01

This paper reports an integrated solution, called BALSA, for the secondary analysis of next generation sequencing data; it exploits the computational power of GPU and an intricate memory management to give a fast and accurate analysis. From raw reads to variants (including SNPs and Indels), BALSA, using just a single computing node with a commodity GPU board, takes 5.5 h to process 50-fold whole genome sequencing (∼750 million 100 bp paired-end reads), or just 25 min for 210-fold whole exome sequencing. BALSA's speed is rooted at its parallel algorithms to effectively exploit a GPU to speed up processes like alignment, realignment and statistical testing. BALSA incorporates a 16-genotype model to support the calling of SNPs and Indels and achieves competitive variant calling accuracy and sensitivity when compared to the ensemble of six popular variant callers. BALSA also supports efficient identification of somatic SNVs and CNVs; experiments showed that BALSA recovers all the previously validated somatic SNVs and CNVs, and it is more sensitive for somatic Indel detection. BALSA outputs variants in VCF format. A pileup-like SNAPSHOT format, while maintaining the same fidelity as BAM in variant calling, enables efficient storage and indexing, and facilitates the App development of downstream analyses. BALSA is available at: http://sourceforge.net/p/balsa.

Efficient Parallel Video Processing Techniques on GPU: From Framework to Implementation

PubMed Central

Su, Huayou; Wen, Mei; Wu, Nan; Ren, Ju; Zhang, Chunyuan

2014-01-01

Through reorganizing the execution order and optimizing the data structure, we proposed an efficient parallel framework for H.264/AVC encoder based on massively parallel architecture. We implemented the proposed framework by CUDA on NVIDIA's GPU. Not only the compute intensive components of the H.264 encoder are parallelized but also the control intensive components are realized effectively, such as CAVLC and deblocking filter. In addition, we proposed serial optimization methods, including the multiresolution multiwindow for motion estimation, multilevel parallel strategy to enhance the parallelism of intracoding as much as possible, component-based parallel CAVLC, and direction-priority deblocking filter. More than 96% of workload of H.264 encoder is offloaded to GPU. Experimental results show that the parallel implementation outperforms the serial program by 20 times of speedup ratio and satisfies the requirement of the real-time HD encoding of 30 fps. The loss of PSNR is from 0.14 dB to 0.77 dB, when keeping the same bitrate. Through the analysis to the kernels, we found that speedup ratios of the compute intensive algorithms are proportional with the computation power of the GPU. However, the performance of the control intensive parts (CAVLC) is much related to the memory bandwidth, which gives an insight for new architecture design. PMID:24757432

A parallel finite element procedure for contact-impact problems using edge-based smooth triangular element and GPU

NASA Astrophysics Data System (ADS)

Cai, Yong; Cui, Xiangyang; Li, Guangyao; Liu, Wenyang

2018-04-01

The edge-smooth finite element method (ES-FEM) can improve the computational accuracy of triangular shell elements and the mesh partition efficiency of complex models. In this paper, an approach is developed to perform explicit finite element simulations of contact-impact problems with a graphical processing unit (GPU) using a special edge-smooth triangular shell element based on ES-FEM. Of critical importance for this problem is achieving finer-grained parallelism to enable efficient data loading and to minimize communication between the device and host. Four kinds of parallel strategies are then developed to efficiently solve these ES-FEM based shell element formulas, and various optimization methods are adopted to ensure aligned memory access. Special focus is dedicated to developing an approach for the parallel construction of edge systems. A parallel hierarchy-territory contact-searching algorithm (HITA) and a parallel penalty function calculation method are embedded in this parallel explicit algorithm. Finally, the program flow is well designed, and a GPU-based simulation system is developed, using Nvidia's CUDA. Several numerical examples are presented to illustrate the high quality of the results obtained with the proposed methods. In addition, the GPU-based parallel computation is shown to significantly reduce the computing time.

MHD code using multi graphical processing units: SMAUG+

NASA Astrophysics Data System (ADS)

Gyenge, N.; Griffiths, M. K.; Erdélyi, R.

2018-01-01

This paper introduces the Sheffield Magnetohydrodynamics Algorithm Using GPUs (SMAUG+), an advanced numerical code for solving magnetohydrodynamic (MHD) problems, using multi-GPU systems. Multi-GPU systems facilitate the development of accelerated codes and enable us to investigate larger model sizes and/or more detailed computational domain resolutions. This is a significant advancement over the parent single-GPU MHD code, SMAUG (Griffiths et al., 2015). Here, we demonstrate the validity of the SMAUG + code, describe the parallelisation techniques and investigate performance benchmarks. The initial configuration of the Orszag-Tang vortex simulations are distributed among 4, 16, 64 and 100 GPUs. Furthermore, different simulation box resolutions are applied: 1000 × 1000, 2044 × 2044, 4000 × 4000 and 8000 × 8000 . We also tested the code with the Brio-Wu shock tube simulations with model size of 800 employing up to 10 GPUs. Based on the test results, we observed speed ups and slow downs, depending on the granularity and the communication overhead of certain parallel tasks. The main aim of the code development is to provide massively parallel code without the memory limitation of a single GPU. By using our code, the applied model size could be significantly increased. We demonstrate that we are able to successfully compute numerically valid and large 2D MHD problems.

Fast Analysis of Molecular Dynamics Trajectories with Graphics Processing Units—Radial Distribution Function Histogramming

PubMed Central

Stone, John E.; Kohlmeyer, Axel

2011-01-01

The calculation of radial distribution functions (RDFs) from molecular dynamics trajectory data is a common and computationally expensive analysis task. The rate limiting step in the calculation of the RDF is building a histogram of the distance between atom pairs in each trajectory frame. Here we present an implementation of this histogramming scheme for multiple graphics processing units (GPUs). The algorithm features a tiling scheme to maximize the reuse of data at the fastest levels of the GPU’s memory hierarchy and dynamic load balancing to allow high performance on heterogeneous configurations of GPUs. Several versions of the RDF algorithm are presented, utilizing the specific hardware features found on different generations of GPUs. We take advantage of larger shared memory and atomic memory operations available on state-of-the-art GPUs to accelerate the code significantly. The use of atomic memory operations allows the fast, limited-capacity on-chip memory to be used much more efficiently, resulting in a fivefold increase in performance compared to the version of the algorithm without atomic operations. The ultimate version of the algorithm running in parallel on four NVIDIA GeForce GTX 480 (Fermi) GPUs was found to be 92 times faster than a multithreaded implementation running on an Intel Xeon 5550 CPU. On this multi-GPU hardware, the RDF between two selections of 1,000,000 atoms each can be calculated in 26.9 seconds per frame. The multi-GPU RDF algorithms described here are implemented in VMD, a widely used and freely available software package for molecular dynamics visualization and analysis. PMID:21547007

Scalable streaming tools for analyzing N-body simulations: Finding halos and investigating excursion sets in one pass

NASA Astrophysics Data System (ADS)

Ivkin, N.; Liu, Z.; Yang, L. F.; Kumar, S. S.; Lemson, G.; Neyrinck, M.; Szalay, A. S.; Braverman, V.; Budavari, T.

2018-04-01

Cosmological N-body simulations play a vital role in studying models for the evolution of the Universe. To compare to observations and make a scientific inference, statistic analysis on large simulation datasets, e.g., finding halos, obtaining multi-point correlation functions, is crucial. However, traditional in-memory methods for these tasks do not scale to the datasets that are forbiddingly large in modern simulations. Our prior paper (Liu et al., 2015) proposes memory-efficient streaming algorithms that can find the largest halos in a simulation with up to 109 particles on a small server or desktop. However, this approach fails when directly scaling to larger datasets. This paper presents a robust streaming tool that leverages state-of-the-art techniques on GPU boosting, sampling, and parallel I/O, to significantly improve performance and scalability. Our rigorous analysis of the sketch parameters improves the previous results from finding the centers of the 103 largest halos (Liu et al., 2015) to ∼ 104 - 105, and reveals the trade-offs between memory, running time and number of halos. Our experiments show that our tool can scale to datasets with up to ∼ 1012 particles while using less than an hour of running time on a single GPU Nvidia GTX 1080.

MOIL-opt: Energy-Conserving Molecular Dynamics on a GPU/CPU system

PubMed Central

Ruymgaart, A. Peter; Cardenas, Alfredo E.; Elber, Ron

2011-01-01

We report an optimized version of the molecular dynamics program MOIL that runs on a shared memory system with OpenMP and exploits the power of a Graphics Processing Unit (GPU). The model is of heterogeneous computing system on a single node with several cores sharing the same memory and a GPU. This is a typical laboratory tool, which provides excellent performance at minimal cost. Besides performance, emphasis is made on accuracy and stability of the algorithm probed by energy conservation for explicit-solvent atomically-detailed-models. Especially for long simulations energy conservation is critical due to the phenomenon known as “energy drift” in which energy errors accumulate linearly as a function of simulation time. To achieve long time dynamics with acceptable accuracy the drift must be particularly small. We identify several means of controlling long-time numerical accuracy while maintaining excellent speedup. To maintain a high level of energy conservation SHAKE and the Ewald reciprocal summation are run in double precision. Double precision summation of real-space non-bonded interactions improves energy conservation. In our best option, the energy drift using 1fs for a time step while constraining the distances of all bonds, is undetectable in 10ns simulation of solvated DHFR (Dihydrofolate reductase). Faster options, shaking only bonds with hydrogen atoms, are also very well behaved and have drifts of less than 1kcal/mol per nanosecond of the same system. CPU/GPU implementations require changes in programming models. We consider the use of a list of neighbors and quadratic versus linear interpolation in lookup tables of different sizes. Quadratic interpolation with a smaller number of grid points is faster than linear lookup tables (with finer representation) without loss of accuracy. Atomic neighbor lists were found most efficient. Typical speedups are about a factor of 10 compared to a single-core single-precision code. PMID:22328867

A Hybrid Task Graph Scheduler for High Performance Image Processing Workflows.

PubMed

Blattner, Timothy; Keyrouz, Walid; Bhattacharyya, Shuvra S; Halem, Milton; Brady, Mary

2017-12-01

Designing applications for scalability is key to improving their performance in hybrid and cluster computing. Scheduling code to utilize parallelism is difficult, particularly when dealing with data dependencies, memory management, data motion, and processor occupancy. The Hybrid Task Graph Scheduler (HTGS) improves programmer productivity when implementing hybrid workflows for multi-core and multi-GPU systems. The Hybrid Task Graph Scheduler (HTGS) is an abstract execution model, framework, and API that increases programmer productivity when implementing hybrid workflows for such systems. HTGS manages dependencies between tasks, represents CPU and GPU memories independently, overlaps computations with disk I/O and memory transfers, keeps multiple GPUs occupied, and uses all available compute resources. Through these abstractions, data motion and memory are explicit; this makes data locality decisions more accessible. To demonstrate the HTGS application program interface (API), we present implementations of two example algorithms: (1) a matrix multiplication that shows how easily task graphs can be used; and (2) a hybrid implementation of microscopy image stitching that reduces code size by ≈ 43% compared to a manually coded hybrid workflow implementation and showcases the minimal overhead of task graphs in HTGS. Both of the HTGS-based implementations show good performance. In image stitching the HTGS implementation achieves similar performance to the hybrid workflow implementation. Matrix multiplication with HTGS achieves 1.3× and 1.8× speedup over the multi-threaded OpenBLAS library for 16k × 16k and 32k × 32k size matrices, respectively.

Multi-GPU configuration of 4D intensity modulated radiation therapy inverse planning using global optimization

NASA Astrophysics Data System (ADS)

Hagan, Aaron; Sawant, Amit; Folkerts, Michael; Modiri, Arezoo

2018-01-01

We report on the design, implementation and characterization of a multi-graphic processing unit (GPU) computational platform for higher-order optimization in radiotherapy treatment planning. In collaboration with a commercial vendor (Varian Medical Systems, Palo Alto, CA), a research prototype GPU-enabled Eclipse (V13.6) workstation was configured. The hardware consisted of dual 8-core Xeon processors, 256 GB RAM and four NVIDIA Tesla K80 general purpose GPUs. We demonstrate the utility of this platform for large radiotherapy optimization problems through the development and characterization of a parallelized particle swarm optimization (PSO) four dimensional (4D) intensity modulated radiation therapy (IMRT) technique. The PSO engine was coupled to the Eclipse treatment planning system via a vendor-provided scripting interface. Specific challenges addressed in this implementation were (i) data management and (ii) non-uniform memory access (NUMA). For the former, we alternated between parameters over which the computation process was parallelized. For the latter, we reduced the amount of data required to be transferred over the NUMA bridge. The datasets examined in this study were approximately 300 GB in size, including 4D computed tomography images, anatomical structure contours and dose deposition matrices. For evaluation, we created a 4D-IMRT treatment plan for one lung cancer patient and analyzed computation speed while varying several parameters (number of respiratory phases, GPUs, PSO particles, and data matrix sizes). The optimized 4D-IMRT plan enhanced sparing of organs at risk by an average reduction of 26% in maximum dose, compared to the clinical optimized IMRT plan, where the internal target volume was used. We validated our computation time analyses in two additional cases. The computation speed in our implementation did not monotonically increase with the number of GPUs. The optimal number of GPUs (five, in our study) is directly related to the hardware specifications. The optimization process took 35 min using 50 PSO particles, 25 iterations and 5 GPUs.

Multi-GPU configuration of 4D intensity modulated radiation therapy inverse planning using global optimization.

PubMed

Hagan, Aaron; Sawant, Amit; Folkerts, Michael; Modiri, Arezoo

2018-01-16

We report on the design, implementation and characterization of a multi-graphic processing unit (GPU) computational platform for higher-order optimization in radiotherapy treatment planning. In collaboration with a commercial vendor (Varian Medical Systems, Palo Alto, CA), a research prototype GPU-enabled Eclipse (V13.6) workstation was configured. The hardware consisted of dual 8-core Xeon processors, 256 GB RAM and four NVIDIA Tesla K80 general purpose GPUs. We demonstrate the utility of this platform for large radiotherapy optimization problems through the development and characterization of a parallelized particle swarm optimization (PSO) four dimensional (4D) intensity modulated radiation therapy (IMRT) technique. The PSO engine was coupled to the Eclipse treatment planning system via a vendor-provided scripting interface. Specific challenges addressed in this implementation were (i) data management and (ii) non-uniform memory access (NUMA). For the former, we alternated between parameters over which the computation process was parallelized. For the latter, we reduced the amount of data required to be transferred over the NUMA bridge. The datasets examined in this study were approximately 300 GB in size, including 4D computed tomography images, anatomical structure contours and dose deposition matrices. For evaluation, we created a 4D-IMRT treatment plan for one lung cancer patient and analyzed computation speed while varying several parameters (number of respiratory phases, GPUs, PSO particles, and data matrix sizes). The optimized 4D-IMRT plan enhanced sparing of organs at risk by an average reduction of [Formula: see text] in maximum dose, compared to the clinical optimized IMRT plan, where the internal target volume was used. We validated our computation time analyses in two additional cases. The computation speed in our implementation did not monotonically increase with the number of GPUs. The optimal number of GPUs (five, in our study) is directly related to the hardware specifications. The optimization process took 35 min using 50 PSO particles, 25 iterations and 5 GPUs.

An MPI + $X$ implementation of contact global search using Kokkos

DOE PAGES

Hansen, Glen A.; Xavier, Patrick G.; Mish, Sam P.; ...

2015-10-05

This paper describes an approach that seeks to parallelize the spatial search associated with computational contact mechanics. In contact mechanics, the purpose of the spatial search is to find “nearest neighbors,” which is the prelude to an imprinting search that resolves the interactions between the external surfaces of contacting bodies. In particular, we are interested in the contact global search portion of the spatial search associated with this operation on domain-decomposition-based meshes. Specifically, we describe an implementation that combines standard domain-decomposition-based MPI-parallel spatial search with thread-level parallelism (MPI-X) available on advanced computer architectures (those with GPU coprocessors). Our goal ismore » to demonstrate the efficacy of the MPI-X paradigm in the overall contact search. Standard MPI-parallel implementations typically use a domain decomposition of the external surfaces of bodies within the domain in an attempt to efficiently distribute computational work. This decomposition may or may not be the same as the volume decomposition associated with the host physics. The parallel contact global search phase is then employed to find and distribute surface entities (nodes and faces) that are needed to compute contact constraints between entities owned by different MPI ranks without further inter-rank communication. Key steps of the contact global search include computing bounding boxes, building surface entity (node and face) search trees and finding and distributing entities required to complete on-rank (local) spatial searches. To enable source-code portability and performance across a variety of different computer architectures, we implemented the algorithm using the Kokkos hardware abstraction library. While we targeted development towards machines with a GPU accelerator per MPI rank, we also report performance results for OpenMP with a conventional multi-core compute node per rank. Results here demonstrate a 47 % decrease in the time spent within the global search algorithm, comparing the reference ACME algorithm with the GPU implementation, on an 18M face problem using four MPI ranks. As a result, while further work remains to maximize performance on the GPU, this result illustrates the potential of the proposed implementation.« less

Performance of Point and Range Queries for In-memory Databases using Radix Trees on GPUs

DOE Office of Scientific and Technical Information (OSTI.GOV)

Alam, Maksudul; Yoginath, Srikanth B; Perumalla, Kalyan S

In in-memory database systems augmented by hardware accelerators, accelerating the index searching operations can greatly increase the runtime performance of database queries. Recently, adaptive radix trees (ART) have been shown to provide very fast index search implementation on the CPU. Here, we focus on an accelerator-based implementation of ART. We present a detailed performance study of our GPU-based adaptive radix tree (GRT) implementation over a variety of key distributions, synthetic benchmarks, and actual keys from music and book data sets. The performance is also compared with other index-searching schemes on the GPU. GRT on modern GPUs achieves some of themore » highest rates of index searches reported in the literature. For point queries, a throughput of up to 106 million and 130 million lookups per second is achieved for sparse and dense keys, respectively. For range queries, GRT yields 600 million and 1000 million lookups per second for sparse and dense keys, respectively, on a large dataset of 64 million 32-bit keys.« less

A GPU-Based Architecture for Real-Time Data Assessment at Synchrotron Experiments

NASA Astrophysics Data System (ADS)

Chilingaryan, Suren; Mirone, Alessandro; Hammersley, Andrew; Ferrero, Claudio; Helfen, Lukas; Kopmann, Andreas; Rolo, Tomy dos Santos; Vagovic, Patrik

2011-08-01

Advances in digital detector technology leads presently to rapidly increasing data rates in imaging experiments. Using fast two-dimensional detectors in computed tomography, the data acquisition can be much faster than the reconstruction if no adequate measures are taken, especially when a high photon flux at synchrotron sources is used. We have optimized the reconstruction software employed at the micro-tomography beamlines of our synchrotron facilities to use the computational power of modern graphic cards. The main paradigm of our approach is the full utilization of all system resources. We use a pipelined architecture, where the GPUs are used as compute coprocessors to reconstruct slices, while the CPUs are preparing the next ones. Special attention is devoted to minimize data transfers between the host and GPU memory and to execute memory transfers in parallel with the computations. We were able to reduce the reconstruction time by a factor 30 and process a typical data set of 20 GB in 40 seconds. The time needed for the first evaluation of the reconstructed sample is reduced significantly and quasi real-time visualization is now possible.

An efficient implementation of 3D high-resolution imaging for large-scale seismic data with GPU/CPU heterogeneous parallel computing

NASA Astrophysics Data System (ADS)

Xu, Jincheng; Liu, Wei; Wang, Jin; Liu, Linong; Zhang, Jianfeng

2018-02-01

De-absorption pre-stack time migration (QPSTM) compensates for the absorption and dispersion of seismic waves by introducing an effective Q parameter, thereby making it an effective tool for 3D, high-resolution imaging of seismic data. Although the optimal aperture obtained via stationary-phase migration reduces the computational cost of 3D QPSTM and yields 3D stationary-phase QPSTM, the associated computational efficiency is still the main problem in the processing of 3D, high-resolution images for real large-scale seismic data. In the current paper, we proposed a division method for large-scale, 3D seismic data to optimize the performance of stationary-phase QPSTM on clusters of graphics processing units (GPU). Then, we designed an imaging point parallel strategy to achieve an optimal parallel computing performance. Afterward, we adopted an asynchronous double buffering scheme for multi-stream to perform the GPU/CPU parallel computing. Moreover, several key optimization strategies of computation and storage based on the compute unified device architecture (CUDA) were adopted to accelerate the 3D stationary-phase QPSTM algorithm. Compared with the initial GPU code, the implementation of the key optimization steps, including thread optimization, shared memory optimization, register optimization and special function units (SFU), greatly improved the efficiency. A numerical example employing real large-scale, 3D seismic data showed that our scheme is nearly 80 times faster than the CPU-QPSTM algorithm. Our GPU/CPU heterogeneous parallel computing framework significant reduces the computational cost and facilitates 3D high-resolution imaging for large-scale seismic data.

Sop-GPU: accelerating biomolecular simulations in the centisecond timescale using graphics processors.

PubMed

Zhmurov, A; Dima, R I; Kholodov, Y; Barsegov, V

2010-11-01

Theoretical exploration of fundamental biological processes involving the forced unraveling of multimeric proteins, the sliding motion in protein fibers and the mechanical deformation of biomolecular assemblies under physiological force loads is challenging even for distributed computing systems. Using a C(α)-based coarse-grained self organized polymer (SOP) model, we implemented the Langevin simulations of proteins on graphics processing units (SOP-GPU program). We assessed the computational performance of an end-to-end application of the program, where all the steps of the algorithm are running on a GPU, by profiling the simulation time and memory usage for a number of test systems. The ∼90-fold computational speedup on a GPU, compared with an optimized central processing unit program, enabled us to follow the dynamics in the centisecond timescale, and to obtain the force-extension profiles using experimental pulling speeds (v(f) = 1-10 μm/s) employed in atomic force microscopy and in optical tweezers-based dynamic force spectroscopy. We found that the mechanical molecular response critically depends on the conditions of force application and that the kinetics and pathways for unfolding change drastically even upon a modest 10-fold increase in v(f). This implies that, to resolve accurately the free energy landscape and to relate the results of single-molecule experiments in vitro and in silico, molecular simulations should be carried out under the experimentally relevant force loads. This can be accomplished in reasonable wall-clock time for biomolecules of size as large as 10(5) residues using the SOP-GPU package. © 2010 Wiley-Liss, Inc.

GPU acceleration towards real-time image reconstruction in 3D tomographic diffractive microscopy

NASA Astrophysics Data System (ADS)

Bailleul, J.; Simon, B.; Debailleul, M.; Liu, H.; Haeberlé, O.

2012-06-01

Phase microscopy techniques regained interest in allowing for the observation of unprepared specimens with excellent temporal resolution. Tomographic diffractive microscopy is an extension of holographic microscopy which permits 3D observations with a finer resolution than incoherent light microscopes. Specimens are imaged by a series of 2D holograms: their accumulation progressively fills the range of frequencies of the specimen in Fourier space. A 3D inverse FFT eventually provides a spatial image of the specimen. Consequently, acquisition then reconstruction are mandatory to produce an image that could prelude real-time control of the observed specimen. The MIPS Laboratory has built a tomographic diffractive microscope with an unsurpassed 130nm resolution but a low imaging speed - no less than one minute. Afterwards, a high-end PC reconstructs the 3D image in 20 seconds. We now expect an interactive system providing preview images during the acquisition for monitoring purposes. We first present a prototype implementing this solution on CPU: acquisition and reconstruction are tied in a producer-consumer scheme, sharing common data into CPU memory. Then we present a prototype dispatching some reconstruction tasks to GPU in order to take advantage of SIMDparallelization for FFT and higher bandwidth for filtering operations. The CPU scheme takes 6 seconds for a 3D image update while the GPU scheme can go down to 2 or > 1 seconds depending on the GPU class. This opens opportunities for 4D imaging of living organisms or crystallization processes. We also consider the relevance of GPU for 3D image interaction in our specific conditions.

Extending the BEAGLE library to a multi-FPGA platform

PubMed Central

2013-01-01

Background Maximum Likelihood (ML)-based phylogenetic inference using Felsenstein’s pruning algorithm is a standard method for estimating the evolutionary relationships amongst a set of species based on DNA sequence data, and is used in popular applications such as RAxML, PHYLIP, GARLI, BEAST, and MrBayes. The Phylogenetic Likelihood Function (PLF) and its associated scaling and normalization steps comprise the computational kernel for these tools. These computations are data intensive but contain fine grain parallelism that can be exploited by coprocessor architectures such as FPGAs and GPUs. A general purpose API called BEAGLE has recently been developed that includes optimized implementations of Felsenstein’s pruning algorithm for various data parallel architectures. In this paper, we extend the BEAGLE API to a multiple Field Programmable Gate Array (FPGA)-based platform called the Convey HC-1. Results The core calculation of our implementation, which includes both the phylogenetic likelihood function (PLF) and the tree likelihood calculation, has an arithmetic intensity of 130 floating-point operations per 64 bytes of I/O, or 2.03 ops/byte. Its performance can thus be calculated as a function of the host platform’s peak memory bandwidth and the implementation’s memory efficiency, as 2.03 × peak bandwidth × memory efficiency. Our FPGA-based platform has a peak bandwidth of 76.8 GB/s and our implementation achieves a memory efficiency of approximately 50%, which gives an average throughput of 78 Gflops. This represents a ~40X speedup when compared with BEAGLE’s CPU implementation on a dual Xeon 5520 and 3X speedup versus BEAGLE’s GPU implementation on a Tesla T10 GPU for very large data sizes. The power consumption is 92 W, yielding a power efficiency of 1.7 Gflops per Watt. Conclusions The use of data parallel architectures to achieve high performance for likelihood-based phylogenetic inference requires high memory bandwidth and a design methodology that emphasizes high memory efficiency. To achieve this objective, we integrated 32 pipelined processing elements (PEs) across four FPGAs. For the design of each PE, we developed a specialized synthesis tool to generate a floating-point pipeline with resource and throughput constraints to match the target platform. We have found that using low-latency floating-point operators can significantly reduce FPGA area and still meet timing requirement on the target platform. We found that this design methodology can achieve performance that exceeds that of a GPU-based coprocessor. PMID:23331707

Introduction of Parallel GPGPU Acceleration Algorithms for the Solution of Radiative Transfer

NASA Technical Reports Server (NTRS)

Godoy, William F.; Liu, Xu

2011-01-01

General-purpose computing on graphics processing units (GPGPU) is a recent technique that allows the parallel graphics processing unit (GPU) to accelerate calculations performed sequentially by the central processing unit (CPU). To introduce GPGPU to radiative transfer, the Gauss-Seidel solution of the well-known expressions for 1-D and 3-D homogeneous, isotropic media is selected as a test case. Different algorithms are introduced to balance memory and GPU-CPU communication, critical aspects of GPGPU. Results show that speed-ups of one to two orders of magnitude are obtained when compared to sequential solutions. The underlying value of GPGPU is its potential extension in radiative solvers (e.g., Monte Carlo, discrete ordinates) at a minimal learning curve.

A GPU-based mipmapping method for water surface visualization

NASA Astrophysics Data System (ADS)

Li, Hua; Quan, Wei; Xu, Chao; Wu, Yan

2018-03-01

Visualization of water surface is a hot topic in computer graphics. In this paper, we presented a fast method to generate wide range of water surface with good image quality both near and far from the viewpoint. This method utilized uniform mesh and Fractal Perlin noise to model water surface. Mipmapping technology was enforced to the surface textures, which adjust the resolution with respect to the distance from the viewpoint and reduce the computing cost. Lighting effect was computed based on shadow mapping technology, Snell's law and Fresnel term. The render pipeline utilizes a CPU-GPU shared memory structure, which improves the rendering efficiency. Experiment results show that our approach visualizes water surface with good image quality at real-time frame rates performance.

Community Detection on the GPU

DOE Office of Scientific and Technical Information (OSTI.GOV)

Naim, Md; Manne, Fredrik; Halappanavar, Mahantesh

We present and evaluate a new GPU algorithm based on the Louvain method for community detection. Our algorithm is the first for this problem that parallelizes the access to individual edges. In this way we can fine tune the load balance when processing networks with nodes of highly varying degrees. This is achieved by scaling the number of threads assigned to each node according to its degree. Extensive experiments show that we obtain speedups up to a factor of 270 compared to the sequential algorithm. The algorithm consistently outperforms other recent shared memory implementations and is only one order ofmore » magnitude slower than the current fastest parallel Louvain method running on a Blue Gene/Q supercomputer using more than 500K threads.« less

Efficient development of memory bounded geo-applications to scale on modern supercomputers

NASA Astrophysics Data System (ADS)

Räss, Ludovic; Omlin, Samuel; Licul, Aleksandar; Podladchikov, Yuri; Herman, Frédéric

2016-04-01

Numerical modeling is an actual key tool in the area of geosciences. The current challenge is to solve problems that are multi-physics and for which the length scale and the place of occurrence might not be known in advance. Also, the spatial extend of the investigated domain might strongly vary in size, ranging from millimeters for reactive transport to kilometers for glacier erosion dynamics. An efficient way to proceed is to develop simple but robust algorithms that perform well and scale on modern supercomputers and permit therefore very high-resolution simulations. We propose an efficient approach to solve memory bounded real-world applications on modern supercomputers architectures. We optimize the software to run on our newly acquired state-of-the-art GPU cluster "octopus". Our approach shows promising preliminary results on important geodynamical and geomechanical problematics: we have developed a Stokes solver for glacier flow and a poromechanical solver including complex rheologies for nonlinear waves in stressed rocks porous rocks. We solve the system of partial differential equations on a regular Cartesian grid and use an iterative finite difference scheme with preconditioning of the residuals. The MPI communication happens only locally (point-to-point); this method is known to scale linearly by construction. The "octopus" GPU cluster, which we use for the computations, has been designed to achieve maximal data transfer throughput at minimal hardware cost. It is composed of twenty compute nodes, each hosting four Nvidia Titan X GPU accelerators. These high-density nodes are interconnected with a parallel (dual-rail) FDR InfiniBand network. Our efforts show promising preliminary results for the different physics investigated. The glacier flow solver achieves good accuracy in the relevant benchmarks and the coupled poromechanical solver permits to explain previously unresolvable focused fluid flow as a natural outcome of the porosity setup. In both cases, near peak memory bandwidth transfer is achieved. Our approach allows us to get the best out of the current hardware.

Global magnetohydrodynamic simulations on multiple GPUs

NASA Astrophysics Data System (ADS)

Wong, Un-Hong; Wong, Hon-Cheng; Ma, Yonghui

2014-01-01

Global magnetohydrodynamic (MHD) models play the major role in investigating the solar wind-magnetosphere interaction. However, the huge computation requirement in global MHD simulations is also the main problem that needs to be solved. With the recent development of modern graphics processing units (GPUs) and the Compute Unified Device Architecture (CUDA), it is possible to perform global MHD simulations in a more efficient manner. In this paper, we present a global magnetohydrodynamic (MHD) simulator on multiple GPUs using CUDA 4.0 with GPUDirect 2.0. Our implementation is based on the modified leapfrog scheme, which is a combination of the leapfrog scheme and the two-step Lax-Wendroff scheme. GPUDirect 2.0 is used in our implementation to drive multiple GPUs. All data transferring and kernel processing are managed with CUDA 4.0 API instead of using MPI or OpenMP. Performance measurements are made on a multi-GPU system with eight NVIDIA Tesla M2050 (Fermi architecture) graphics cards. These measurements show that our multi-GPU implementation achieves a peak performance of 97.36 GFLOPS in double precision.

Matrix Algebra for GPU and Multicore Architectures (MAGMA) for Large Petascale Systems

DOE Office of Scientific and Technical Information (OSTI.GOV)

Dongarra, Jack J.; Tomov, Stanimire

2014-03-24

The goal of the MAGMA project is to create a new generation of linear algebra libraries that achieve the fastest possible time to an accurate solution on hybrid Multicore+GPU-based systems, using all the processing power that future high-end systems can make available within given energy constraints. Our efforts at the University of Tennessee achieved the goals set in all of the five areas identified in the proposal: 1. Communication optimal algorithms; 2. Autotuning for GPU and hybrid processors; 3. Scheduling and memory management techniques for heterogeneity and scale; 4. Fault tolerance and robustness for large scale systems; 5. Building energymore » efficiency into software foundations. The University of Tennessee’s main contributions, as proposed, were the research and software development of new algorithms for hybrid multi/many-core CPUs and GPUs, as related to two-sided factorizations and complete eigenproblem solvers, hybrid BLAS, and energy efficiency for dense, as well as sparse, operations. Furthermore, as proposed, we investigated and experimented with various techniques targeting the five main areas outlined.« less

Acceleration of discrete stochastic biochemical simulation using GPGPU.

PubMed

Sumiyoshi, Kei; Hirata, Kazuki; Hiroi, Noriko; Funahashi, Akira

2015-01-01

For systems made up of a small number of molecules, such as a biochemical network in a single cell, a simulation requires a stochastic approach, instead of a deterministic approach. The stochastic simulation algorithm (SSA) simulates the stochastic behavior of a spatially homogeneous system. Since stochastic approaches produce different results each time they are used, multiple runs are required in order to obtain statistical results; this results in a large computational cost. We have implemented a parallel method for using SSA to simulate a stochastic model; the method uses a graphics processing unit (GPU), which enables multiple realizations at the same time, and thus reduces the computational time and cost. During the simulation, for the purpose of analysis, each time course is recorded at each time step. A straightforward implementation of this method on a GPU is about 16 times faster than a sequential simulation on a CPU with hybrid parallelization; each of the multiple simulations is run simultaneously, and the computational tasks within each simulation are parallelized. We also implemented an improvement to the memory access and reduced the memory footprint, in order to optimize the computations on the GPU. We also implemented an asynchronous data transfer scheme to accelerate the time course recording function. To analyze the acceleration of our implementation on various sizes of model, we performed SSA simulations on different model sizes and compared these computation times to those for sequential simulations with a CPU. When used with the improved time course recording function, our method was shown to accelerate the SSA simulation by a factor of up to 130.

Acceleration of discrete stochastic biochemical simulation using GPGPU

PubMed Central

Sumiyoshi, Kei; Hirata, Kazuki; Hiroi, Noriko; Funahashi, Akira

2015-01-01

For systems made up of a small number of molecules, such as a biochemical network in a single cell, a simulation requires a stochastic approach, instead of a deterministic approach. The stochastic simulation algorithm (SSA) simulates the stochastic behavior of a spatially homogeneous system. Since stochastic approaches produce different results each time they are used, multiple runs are required in order to obtain statistical results; this results in a large computational cost. We have implemented a parallel method for using SSA to simulate a stochastic model; the method uses a graphics processing unit (GPU), which enables multiple realizations at the same time, and thus reduces the computational time and cost. During the simulation, for the purpose of analysis, each time course is recorded at each time step. A straightforward implementation of this method on a GPU is about 16 times faster than a sequential simulation on a CPU with hybrid parallelization; each of the multiple simulations is run simultaneously, and the computational tasks within each simulation are parallelized. We also implemented an improvement to the memory access and reduced the memory footprint, in order to optimize the computations on the GPU. We also implemented an asynchronous data transfer scheme to accelerate the time course recording function. To analyze the acceleration of our implementation on various sizes of model, we performed SSA simulations on different model sizes and compared these computation times to those for sequential simulations with a CPU. When used with the improved time course recording function, our method was shown to accelerate the SSA simulation by a factor of up to 130. PMID:25762936

Deep Packet/Flow Analysis using GPUs

DOE Office of Scientific and Technical Information (OSTI.GOV)

Gong, Qian; Wu, Wenji; DeMar, Phil

Deep packet inspection (DPI) faces severe performance challenges in high-speed networks (40/100 GE) as it requires a large amount of raw computing power and high I/O throughputs. Recently, researchers have tentatively used GPUs to address the above issues and boost the performance of DPI. Typically, DPI applications involve highly complex operations in both per-packet and per-flow data level, often in real-time. The parallel architecture of GPUs fits exceptionally well for per-packet network traffic processing. However, for stateful network protocols such as TCP, their data stream need to be reconstructed in a per-flow level to deliver a consistent content analysis. Sincemore » the flow-centric operations are naturally antiparallel and often require large memory space for buffering out-of-sequence packets, they can be problematic for GPUs, whose memory is normally limited to several gigabytes. In this work, we present a highly efficient GPU-based deep packet/flow analysis framework. The proposed design includes a purely GPU-implemented flow tracking and TCP stream reassembly. Instead of buffering and waiting for TCP packets to become in sequence, our framework process the packets in batch and uses a deterministic finite automaton (DFA) with prefix-/suffix- tree method to detect patterns across out-of-sequence packets that happen to be located in different batches. In conclusion, evaluation shows that our code can reassemble and forward tens of millions of packets per second and conduct a stateful signature-based deep packet inspection at 55 Gbit/s using an NVIDIA K40 GPU.« less

CUDA programs for the GPU computing of the Swendsen-Wang multi-cluster spin flip algorithm: 2D and 3D Ising, Potts, and XY models

NASA Astrophysics Data System (ADS)

Komura, Yukihiro; Okabe, Yutaka

2014-03-01

We present sample CUDA programs for the GPU computing of the Swendsen-Wang multi-cluster spin flip algorithm. We deal with the classical spin models; the Ising model, the q-state Potts model, and the classical XY model. As for the lattice, both the 2D (square) lattice and the 3D (simple cubic) lattice are treated. We already reported the idea of the GPU implementation for 2D models (Komura and Okabe, 2012). We here explain the details of sample programs, and discuss the performance of the present GPU implementation for the 3D Ising and XY models. We also show the calculated results of the moment ratio for these models, and discuss phase transitions. Catalogue identifier: AERM_v1_0 Program summary URL:http://cpc.cs.qub.ac.uk/summaries/AERM_v1_0.html Program obtainable from: CPC Program Library, Queen’s University, Belfast, N. Ireland Licensing provisions: Standard CPC licence, http://cpc.cs.qub.ac.uk/licence/licence.html No. of lines in distributed program, including test data, etc.: 5632 No. of bytes in distributed program, including test data, etc.: 14688 Distribution format: tar.gz Programming language: C, CUDA. Computer: System with an NVIDIA CUDA enabled GPU. Operating system: System with an NVIDIA CUDA enabled GPU. Classification: 23. External routines: NVIDIA CUDA Toolkit 3.0 or newer Nature of problem: Monte Carlo simulation of classical spin systems. Ising, q-state Potts model, and the classical XY model are treated for both two-dimensional and three-dimensional lattices. Solution method: GPU-based Swendsen-Wang multi-cluster spin flip Monte Carlo method. The CUDA implementation for the cluster-labeling is based on the work by Hawick et al. [1] and that by Kalentev et al. [2]. Restrictions: The system size is limited depending on the memory of a GPU. Running time: For the parameters used in the sample programs, it takes about a minute for each program. Of course, it depends on the system size, the number of Monte Carlo steps, etc. References: [1] K.A. Hawick, A. Leist, and D. P. Playne, Parallel Computing 36 (2010) 655-678 [2] O. Kalentev, A. Rai, S. Kemnitzb, and R. Schneider, J. Parallel Distrib. Comput. 71 (2011) 615-620

Interactive Light Stimulus Generation with High Performance Real-Time Image Processing and Simple Scripting.

PubMed

Szécsi, László; Kacsó, Ágota; Zeck, Günther; Hantz, Péter

2017-01-01

Light stimulation with precise and complex spatial and temporal modulation is demanded by a series of research fields like visual neuroscience, optogenetics, ophthalmology, and visual psychophysics. We developed a user-friendly and flexible stimulus generating framework (GEARS GPU-based Eye And Retina Stimulation Software), which offers access to GPU computing power, and allows interactive modification of stimulus parameters during experiments. Furthermore, it has built-in support for driving external equipment, as well as for synchronization tasks, via USB ports. The use of GEARS does not require elaborate programming skills. The necessary scripting is visually aided by an intuitive interface, while the details of the underlying software and hardware components remain hidden. Internally, the software is a C++/Python hybrid using OpenGL graphics. Computations are performed on the GPU, and are defined in the GLSL shading language. However, all GPU settings, including the GPU shader programs, are automatically generated by GEARS. This is configured through a method encountered in game programming, which allows high flexibility: stimuli are straightforwardly composed using a broad library of basic components. Stimulus rendering is implemented solely in C++, therefore intermediary libraries for interfacing could be omitted. This enables the program to perform computationally demanding tasks like en-masse random number generation or real-time image processing by local and global operations.

A non-voxel-based broad-beam (NVBB) framework for IMRT treatment planning.

PubMed

Lu, Weiguo

2010-12-07

We present a novel framework that enables very large scale intensity-modulated radiation therapy (IMRT) planning in limited computation resources with improvements in cost, plan quality and planning throughput. Current IMRT optimization uses a voxel-based beamlet superposition (VBS) framework that requires pre-calculation and storage of a large amount of beamlet data, resulting in large temporal and spatial complexity. We developed a non-voxel-based broad-beam (NVBB) framework for IMRT capable of direct treatment parameter optimization (DTPO). In this framework, both objective function and derivative are evaluated based on the continuous viewpoint, abandoning 'voxel' and 'beamlet' representations. Thus pre-calculation and storage of beamlets are no longer needed. The NVBB framework has linear complexities (O(N(3))) in both space and time. The low memory, full computation and data parallelization nature of the framework render its efficient implementation on the graphic processing unit (GPU). We implemented the NVBB framework and incorporated it with the TomoTherapy treatment planning system (TPS). The new TPS runs on a single workstation with one GPU card (NVBB-GPU). Extensive verification/validation tests were performed in house and via third parties. Benchmarks on dose accuracy, plan quality and throughput were compared with the commercial TomoTherapy TPS that is based on the VBS framework and uses a computer cluster with 14 nodes (VBS-cluster). For all tests, the dose accuracy of these two TPSs is comparable (within 1%). Plan qualities were comparable with no clinically significant difference for most cases except that superior target uniformity was seen in the NVBB-GPU for some cases. However, the planning time using the NVBB-GPU was reduced many folds over the VBS-cluster. In conclusion, we developed a novel NVBB framework for IMRT optimization. The continuous viewpoint and DTPO nature of the algorithm eliminate the need for beamlets and lead to better plan quality. The computation parallelization on a GPU instead of a computer cluster significantly reduces hardware and service costs. Compared with using the current VBS framework on a computer cluster, the planning time is significantly reduced using the NVBB framework on a single workstation with a GPU card.

Exploiting on-node heterogeneity for in-situ analytics of climate simulations via a functional partitioning framework

NASA Astrophysics Data System (ADS)

Sapra, Karan; Gupta, Saurabh; Atchley, Scott; Anantharaj, Valentine; Miller, Ross; Vazhkudai, Sudharshan

2016-04-01

Efficient resource utilization is critical for improved end-to-end computing and workflow of scientific applications. Heterogeneous node architectures, such as the GPU-enabled Titan supercomputer at the Oak Ridge Leadership Computing Facility (OLCF), present us with further challenges. In many HPC applications on Titan, the accelerators are the primary compute engines while the CPUs orchestrate the offloading of work onto the accelerators, and moving the output back to the main memory. On the other hand, applications that do not exploit GPUs, the CPU usage is dominant while the GPUs idle. We utilized Heterogenous Functional Partitioning (HFP) runtime framework that can optimize usage of resources on a compute node to expedite an application's end-to-end workflow. This approach is different from existing techniques for in-situ analyses in that it provides a framework for on-the-fly analysis on-node by dynamically exploiting under-utilized resources therein. We have implemented in the Community Earth System Model (CESM) a new concurrent diagnostic processing capability enabled by the HFP framework. Various single variate statistics, such as means and distributions, are computed in-situ by launching HFP tasks on the GPU via the node local HFP daemon. Since our current configuration of CESM does not use GPU resources heavily, we can move these tasks to GPU using the HFP framework. Each rank running the atmospheric model in CESM pushes the variables of of interest via HFP function calls to the HFP daemon. This node local daemon is responsible for receiving the data from main program and launching the designated analytics tasks on the GPU. We have implemented these analytics tasks in C and use OpenACC directives to enable GPU acceleration. This methodology is also advantageous while executing GPU-enabled configurations of CESM when the CPUs will be idle during portions of the runtime. In our implementation results, we demonstrate that it is more efficient to use HFP framework to offload the tasks to GPUs instead of doing it in the main application. We observe increased resource utilization and overall productivity in this approach by using HFP framework for end-to-end workflow.

Parallel tempering simulation of the three-dimensional Edwards-Anderson model with compact asynchronous multispin coding on GPU

NASA Astrophysics Data System (ADS)

Fang, Ye; Feng, Sheng; Tam, Ka-Ming; Yun, Zhifeng; Moreno, Juana; Ramanujam, J.; Jarrell, Mark

2014-10-01

Monte Carlo simulations of the Ising model play an important role in the field of computational statistical physics, and they have revealed many properties of the model over the past few decades. However, the effect of frustration due to random disorder, in particular the possible spin glass phase, remains a crucial but poorly understood problem. One of the obstacles in the Monte Carlo simulation of random frustrated systems is their long relaxation time making an efficient parallel implementation on state-of-the-art computation platforms highly desirable. The Graphics Processing Unit (GPU) is such a platform that provides an opportunity to significantly enhance the computational performance and thus gain new insight into this problem. In this paper, we present optimization and tuning approaches for the CUDA implementation of the spin glass simulation on GPUs. We discuss the integration of various design alternatives, such as GPU kernel construction with minimal communication, memory tiling, and look-up tables. We present a binary data format, Compact Asynchronous Multispin Coding (CAMSC), which provides an additional 28.4% speedup compared with the traditionally used Asynchronous Multispin Coding (AMSC). Our overall design sustains a performance of 33.5 ps per spin flip attempt for simulating the three-dimensional Edwards-Anderson model with parallel tempering, which significantly improves the performance over existing GPU implementations.

AMITIS: A 3D GPU-Based Hybrid-PIC Model for Space and Plasma Physics

NASA Astrophysics Data System (ADS)

Fatemi, Shahab; Poppe, Andrew R.; Delory, Gregory T.; Farrell, William M.

2017-05-01

We have developed, for the first time, an advanced modeling infrastructure in space simulations (AMITIS) with an embedded three-dimensional self-consistent grid-based hybrid model of plasma (kinetic ions and fluid electrons) that runs entirely on graphics processing units (GPUs). The model uses NVIDIA GPUs and their associated parallel computing platform, CUDA, developed for general purpose processing on GPUs. The model uses a single CPU-GPU pair, where the CPU transfers data between the system and GPU memory, executes CUDA kernels, and writes simulation outputs on the disk. All computations, including moving particles, calculating macroscopic properties of particles on a grid, and solving hybrid model equations are processed on a single GPU. We explain various computing kernels within AMITIS and compare their performance with an already existing well-tested hybrid model of plasma that runs in parallel using multi-CPU platforms. We show that AMITIS runs ∼10 times faster than the parallel CPU-based hybrid model. We also introduce an implicit solver for computation of Faraday’s Equation, resulting in an explicit-implicit scheme for the hybrid model equation. We show that the proposed scheme is stable and accurate. We examine the AMITIS energy conservation and show that the energy is conserved with an error < 0.2% after 500,000 timesteps, even when a very low number of particles per cell is used.

SU-E-T-558: Monte Carlo Photon Transport Simulations On GPU with Quadric Geometry

DOE Office of Scientific and Technical Information (OSTI.GOV)

Chi, Y; Tian, Z; Jiang, S

Purpose: Monte Carlo simulation on GPU has experienced rapid advancements over the past a few years and tremendous accelerations have been achieved. Yet existing packages were developed only in voxelized geometry. In some applications, e.g. radioactive seed modeling, simulations in more complicated geometry are needed. This abstract reports our initial efforts towards developing a quadric geometry module aiming at expanding the application scope of GPU-based MC simulations. Methods: We defined the simulation geometry consisting of a number of homogeneous bodies, each specified by its material composition and limiting surfaces characterized by quadric functions. A tree data structure was utilized tomore » define geometric relationship between different bodies. We modified our GPU-based photon MC transport package to incorporate this geometry. Specifically, geometry parameters were loaded into GPU’s shared memory for fast access. Geometry functions were rewritten to enable the identification of the body that contains the current particle location via a fast searching algorithm based on the tree data structure. Results: We tested our package in an example problem of HDR-brachytherapy dose calculation for shielded cylinder. The dose under the quadric geometry and that under the voxelized geometry agreed in 94.2% of total voxels within 20% isodose line based on a statistical t-test (95% confidence level), where the reference dose was defined to be the one at 0.5cm away from the cylinder surface. It took 243sec to transport 100million source photons under this quadric geometry on an NVidia Titan GPU card. Compared with simulation time of 99.6sec in the voxelized geometry, including quadric geometry reduced efficiency due to the complicated geometry-related computations. Conclusion: Our GPU-based MC package has been extended to support photon transport simulation in quadric geometry. Satisfactory accuracy was observed with a reduced efficiency. Developments for charged particle transport in this geometry are currently in progress.« less

Performance Analysis of GFDL's GCM Line-By-Line Radiative Transfer Model on GPU and MIC Architectures

NASA Astrophysics Data System (ADS)

Menzel, R.; Paynter, D.; Jones, A. L.

2017-12-01

Due to their relatively low computational cost, radiative transfer models in global climate models (GCMs) run on traditional CPU architectures generally consist of shortwave and longwave parameterizations over a small number of wavelength bands. With the rise of newer GPU and MIC architectures, however, the performance of high resolution line-by-line radiative transfer models may soon approach those of the physical parameterizations currently employed in GCMs. Here we present an analysis of the current performance of a new line-by-line radiative transfer model currently under development at GFDL. Although originally designed to specifically exploit GPU architectures through the use of CUDA, the radiative transfer model has recently been extended to include OpenMP in an effort to also effectively target MIC architectures such as Intel's Xeon Phi. Using input data provided by the upcoming Radiative Forcing Model Intercomparison Project (RFMIP, as part of CMIP 6), we compare model results and performance data for various model configurations and spectral resolutions run on both GPU and Intel Knights Landing architectures to analogous runs of the standard Oxford Reference Forward Model on traditional CPUs.

DOE Office of Scientific and Technical Information (OSTI.GOV)

Li, Lingda; Hayes, Ari; Song, Shuaiwen

Modern GPUs employ cache to improve memory system efficiency. However, large amount of cache space is underutilized due to irregular memory accesses and poor spatial locality which exhibited commonly in GPU applications. Our experiments show that using smaller cache lines could improve cache space utilization, but it also frequently suffers from significant performance loss by introducing large amount of extra cache requests. In this work, we propose a novel cache design named tag-split cache (TSC) that enables fine-grained cache storage to address the problem of cache space underutilization while keeping memory request number unchanged. TSC divides tag into two partsmore » to reduce storage overhead, and it supports multiple cache line replacement in one cycle.« less

Performance analysis of the FDTD method applied to holographic volume gratings: Multi-core CPU versus GPU computing

NASA Astrophysics Data System (ADS)

Francés, J.; Bleda, S.; Neipp, C.; Márquez, A.; Pascual, I.; Beléndez, A.

2013-03-01

The finite-difference time-domain method (FDTD) allows electromagnetic field distribution analysis as a function of time and space. The method is applied to analyze holographic volume gratings (HVGs) for the near-field distribution at optical wavelengths. Usually, this application requires the simulation of wide areas, which implies more memory and time processing. In this work, we propose a specific implementation of the FDTD method including several add-ons for a precise simulation of optical diffractive elements. Values in the near-field region are computed considering the illumination of the grating by means of a plane wave for different angles of incidence and including absorbing boundaries as well. We compare the results obtained by FDTD with those obtained using a matrix method (MM) applied to diffraction gratings. In addition, we have developed two optimized versions of the algorithm, for both CPU and GPU, in order to analyze the improvement of using the new NVIDIA Fermi GPU architecture versus highly tuned multi-core CPU as a function of the size simulation. In particular, the optimized CPU implementation takes advantage of the arithmetic and data transfer streaming SIMD (single instruction multiple data) extensions (SSE) included explicitly in the code and also of multi-threading by means of OpenMP directives. A good agreement between the results obtained using both FDTD and MM methods is obtained, thus validating our methodology. Moreover, the performance of the GPU is compared to the SSE+OpenMP CPU implementation, and it is quantitatively determined that a highly optimized CPU program can be competitive for a wider range of simulation sizes, whereas GPU computing becomes more powerful for large-scale simulations.

Algorithms of GPU-enabled reactive force field (ReaxFF) molecular dynamics.

PubMed

Zheng, Mo; Li, Xiaoxia; Guo, Li

2013-04-01

Reactive force field (ReaxFF), a recent and novel bond order potential, allows for reactive molecular dynamics (ReaxFF MD) simulations for modeling larger and more complex molecular systems involving chemical reactions when compared with computation intensive quantum mechanical methods. However, ReaxFF MD can be approximately 10-50 times slower than classical MD due to its explicit modeling of bond forming and breaking, the dynamic charge equilibration at each time-step, and its one order smaller time-step than the classical MD, all of which pose significant computational challenges in simulation capability to reach spatio-temporal scales of nanometers and nanoseconds. The very recent advances of graphics processing unit (GPU) provide not only highly favorable performance for GPU enabled MD programs compared with CPU implementations but also an opportunity to manage with the computing power and memory demanding nature imposed on computer hardware by ReaxFF MD. In this paper, we present the algorithms of GMD-Reax, the first GPU enabled ReaxFF MD program with significantly improved performance surpassing CPU implementations on desktop workstations. The performance of GMD-Reax has been benchmarked on a PC equipped with a NVIDIA C2050 GPU for coal pyrolysis simulation systems with atoms ranging from 1378 to 27,283. GMD-Reax achieved speedups as high as 12 times faster than Duin et al.'s FORTRAN codes in Lammps on 8 CPU cores and 6 times faster than the Lammps' C codes based on PuReMD in terms of the simulation time per time-step averaged over 100 steps. GMD-Reax could be used as a new and efficient computational tool for exploiting very complex molecular reactions via ReaxFF MD simulation on desktop workstations. Copyright © 2013 Elsevier Inc. All rights reserved.

A nonvoxel-based dose convolution/superposition algorithm optimized for scalable GPU architectures.

PubMed

Neylon, J; Sheng, K; Yu, V; Chen, Q; Low, D A; Kupelian, P; Santhanam, A

2014-10-01

Real-time adaptive planning and treatment has been infeasible due in part to its high computational complexity. There have been many recent efforts to utilize graphics processing units (GPUs) to accelerate the computational performance and dose accuracy in radiation therapy. Data structure and memory access patterns are the key GPU factors that determine the computational performance and accuracy. In this paper, the authors present a nonvoxel-based (NVB) approach to maximize computational and memory access efficiency and throughput on the GPU. The proposed algorithm employs a ray-tracing mechanism to restructure the 3D data sets computed from the CT anatomy into a nonvoxel-based framework. In a process that takes only a few milliseconds of computing time, the algorithm restructured the data sets by ray-tracing through precalculated CT volumes to realign the coordinate system along the convolution direction, as defined by zenithal and azimuthal angles. During the ray-tracing step, the data were resampled according to radial sampling and parallel ray-spacing parameters making the algorithm independent of the original CT resolution. The nonvoxel-based algorithm presented in this paper also demonstrated a trade-off in computational performance and dose accuracy for different coordinate system configurations. In order to find the best balance between the computed speedup and the accuracy, the authors employed an exhaustive parameter search on all sampling parameters that defined the coordinate system configuration: zenithal, azimuthal, and radial sampling of the convolution algorithm, as well as the parallel ray spacing during ray tracing. The angular sampling parameters were varied between 4 and 48 discrete angles, while both radial sampling and parallel ray spacing were varied from 0.5 to 10 mm. The gamma distribution analysis method (γ) was used to compare the dose distributions using 2% and 2 mm dose difference and distance-to-agreement criteria, respectively. Accuracy was investigated using three distinct phantoms with varied geometries and heterogeneities and on a series of 14 segmented lung CT data sets. Performance gains were calculated using three 256 mm cube homogenous water phantoms, with isotropic voxel dimensions of 1, 2, and 4 mm. The nonvoxel-based GPU algorithm was independent of the data size and provided significant computational gains over the CPU algorithm for large CT data sizes. The parameter search analysis also showed that the ray combination of 8 zenithal and 8 azimuthal angles along with 1 mm radial sampling and 2 mm parallel ray spacing maintained dose accuracy with greater than 99% of voxels passing the γ test. Combining the acceleration obtained from GPU parallelization with the sampling optimization, the authors achieved a total performance improvement factor of >175 000 when compared to our voxel-based ground truth CPU benchmark and a factor of 20 compared with a voxel-based GPU dose convolution method. The nonvoxel-based convolution method yielded substantial performance improvements over a generic GPU implementation, while maintaining accuracy as compared to a CPU computed ground truth dose distribution. Such an algorithm can be a key contribution toward developing tools for adaptive radiation therapy systems.

A nonvoxel-based dose convolution/superposition algorithm optimized for scalable GPU architectures

DOE Office of Scientific and Technical Information (OSTI.GOV)

Neylon, J., E-mail: jneylon@mednet.ucla.edu; Sheng, K.; Yu, V.

Purpose: Real-time adaptive planning and treatment has been infeasible due in part to its high computational complexity. There have been many recent efforts to utilize graphics processing units (GPUs) to accelerate the computational performance and dose accuracy in radiation therapy. Data structure and memory access patterns are the key GPU factors that determine the computational performance and accuracy. In this paper, the authors present a nonvoxel-based (NVB) approach to maximize computational and memory access efficiency and throughput on the GPU. Methods: The proposed algorithm employs a ray-tracing mechanism to restructure the 3D data sets computed from the CT anatomy intomore » a nonvoxel-based framework. In a process that takes only a few milliseconds of computing time, the algorithm restructured the data sets by ray-tracing through precalculated CT volumes to realign the coordinate system along the convolution direction, as defined by zenithal and azimuthal angles. During the ray-tracing step, the data were resampled according to radial sampling and parallel ray-spacing parameters making the algorithm independent of the original CT resolution. The nonvoxel-based algorithm presented in this paper also demonstrated a trade-off in computational performance and dose accuracy for different coordinate system configurations. In order to find the best balance between the computed speedup and the accuracy, the authors employed an exhaustive parameter search on all sampling parameters that defined the coordinate system configuration: zenithal, azimuthal, and radial sampling of the convolution algorithm, as well as the parallel ray spacing during ray tracing. The angular sampling parameters were varied between 4 and 48 discrete angles, while both radial sampling and parallel ray spacing were varied from 0.5 to 10 mm. The gamma distribution analysis method (γ) was used to compare the dose distributions using 2% and 2 mm dose difference and distance-to-agreement criteria, respectively. Accuracy was investigated using three distinct phantoms with varied geometries and heterogeneities and on a series of 14 segmented lung CT data sets. Performance gains were calculated using three 256 mm cube homogenous water phantoms, with isotropic voxel dimensions of 1, 2, and 4 mm. Results: The nonvoxel-based GPU algorithm was independent of the data size and provided significant computational gains over the CPU algorithm for large CT data sizes. The parameter search analysis also showed that the ray combination of 8 zenithal and 8 azimuthal angles along with 1 mm radial sampling and 2 mm parallel ray spacing maintained dose accuracy with greater than 99% of voxels passing the γ test. Combining the acceleration obtained from GPU parallelization with the sampling optimization, the authors achieved a total performance improvement factor of >175 000 when compared to our voxel-based ground truth CPU benchmark and a factor of 20 compared with a voxel-based GPU dose convolution method. Conclusions: The nonvoxel-based convolution method yielded substantial performance improvements over a generic GPU implementation, while maintaining accuracy as compared to a CPU computed ground truth dose distribution. Such an algorithm can be a key contribution toward developing tools for adaptive radiation therapy systems.« less

A scalable approach to solving dense linear algebra problems on hybrid CPU-GPU systems

DOE PAGES

Song, Fengguang; Dongarra, Jack

2014-10-01

Aiming to fully exploit the computing power of all CPUs and all graphics processing units (GPUs) on hybrid CPU-GPU systems to solve dense linear algebra problems, in this paper we design a class of heterogeneous tile algorithms to maximize the degree of parallelism, to minimize the communication volume, and to accommodate the heterogeneity between CPUs and GPUs. The new heterogeneous tile algorithms are executed upon our decentralized dynamic scheduling runtime system, which schedules a task graph dynamically and transfers data between compute nodes automatically. The runtime system uses a new distributed task assignment protocol to solve data dependencies between tasksmore » without any coordination between processing units. By overlapping computation and communication through dynamic scheduling, we are able to attain scalable performance for the double-precision Cholesky factorization and QR factorization. Finally, our approach demonstrates a performance comparable to Intel MKL on shared-memory multicore systems and better performance than both vendor (e.g., Intel MKL) and open source libraries (e.g., StarPU) in the following three environments: heterogeneous clusters with GPUs, conventional clusters without GPUs, and shared-memory systems with multiple GPUs.« less

A scalable approach to solving dense linear algebra problems on hybrid CPU-GPU systems

DOE Office of Scientific and Technical Information (OSTI.GOV)

Song, Fengguang; Dongarra, Jack

Aiming to fully exploit the computing power of all CPUs and all graphics processing units (GPUs) on hybrid CPU-GPU systems to solve dense linear algebra problems, in this paper we design a class of heterogeneous tile algorithms to maximize the degree of parallelism, to minimize the communication volume, and to accommodate the heterogeneity between CPUs and GPUs. The new heterogeneous tile algorithms are executed upon our decentralized dynamic scheduling runtime system, which schedules a task graph dynamically and transfers data between compute nodes automatically. The runtime system uses a new distributed task assignment protocol to solve data dependencies between tasksmore » without any coordination between processing units. By overlapping computation and communication through dynamic scheduling, we are able to attain scalable performance for the double-precision Cholesky factorization and QR factorization. Finally, our approach demonstrates a performance comparable to Intel MKL on shared-memory multicore systems and better performance than both vendor (e.g., Intel MKL) and open source libraries (e.g., StarPU) in the following three environments: heterogeneous clusters with GPUs, conventional clusters without GPUs, and shared-memory systems with multiple GPUs.« less

Interactive Light Stimulus Generation with High Performance Real-Time Image Processing and Simple Scripting

PubMed Central

Szécsi, László; Kacsó, Ágota; Zeck, Günther; Hantz, Péter

2017-01-01

Light stimulation with precise and complex spatial and temporal modulation is demanded by a series of research fields like visual neuroscience, optogenetics, ophthalmology, and visual psychophysics. We developed a user-friendly and flexible stimulus generating framework (GEARS GPU-based Eye And Retina Stimulation Software), which offers access to GPU computing power, and allows interactive modification of stimulus parameters during experiments. Furthermore, it has built-in support for driving external equipment, as well as for synchronization tasks, via USB ports. The use of GEARS does not require elaborate programming skills. The necessary scripting is visually aided by an intuitive interface, while the details of the underlying software and hardware components remain hidden. Internally, the software is a C++/Python hybrid using OpenGL graphics. Computations are performed on the GPU, and are defined in the GLSL shading language. However, all GPU settings, including the GPU shader programs, are automatically generated by GEARS. This is configured through a method encountered in game programming, which allows high flexibility: stimuli are straightforwardly composed using a broad library of basic components. Stimulus rendering is implemented solely in C++, therefore intermediary libraries for interfacing could be omitted. This enables the program to perform computationally demanding tasks like en-masse random number generation or real-time image processing by local and global operations. PMID:29326579

GPU Multi-Scale Particle Tracking and Multi-Fluid Simulations of the Radiation Belts

NASA Astrophysics Data System (ADS)

Ziemba, T.; Carscadden, J.; O'Donnell, D.; Winglee, R.; Harnett, E.; Cash, M.

2007-12-01

The properties of the radiation belts can vary dramatically under the influence of magnetic storms and storm-time substorms. The task of understanding and predicting radiation belt properties is made difficult because their properties determined by global processes as well as small-scale wave-particle interactions. A full solution to the problem will require major innovations in technique and computer hardware. The proposed work will demonstrates liked particle tracking codes with new multi-scale/multi-fluid global simulations that provide the first means to include small-scale processes within the global magnetospheric context. A large hurdle to the problem is having sufficient computer hardware that is able to handle the dissipate temporal and spatial scale sizes. A major innovation of the work is that the codes are designed to run of graphics processing units (GPUs). GPUs are intrinsically highly parallelized systems that provide more than an order of magnitude computing speed over a CPU based systems, for little more cost than a high end-workstation. Recent advancements in GPU technologies allow for full IEEE float specifications with performance up to several hundred GFLOPs per GPU and new software architectures have recently become available to ease the transition from graphics based to scientific applications. This allows for a cheap alternative to standard supercomputing methods and should increase the time to discovery. A demonstration of the code pushing more than 500,000 particles faster than real time is presented, and used to provide new insight into radiation belt dynamics.

Online Tracking Algorithms on GPUs for the P̅ANDA Experiment at FAIR

NASA Astrophysics Data System (ADS)

Bianchi, L.; Herten, A.; Ritman, J.; Stockmanns, T.; Adinetz, A.; Kraus, J.; Pleiter, D.

2015-12-01

P̅ANDA is a future hadron and nuclear physics experiment at the FAIR facility in construction in Darmstadt, Germany. In contrast to the majority of current experiments, PANDA's strategy for data acquisition is based on event reconstruction from free-streaming data, performed in real time entirely by software algorithms using global detector information. This paper reports the status of the development of algorithms for the reconstruction of charged particle tracks, optimized online data processing applications, using General-Purpose Graphic Processing Units (GPU). Two algorithms for trackfinding, the Triplet Finder and the Circle Hough, are described, and details of their GPU implementations are highlighted. Average track reconstruction times of less than 100 ns are obtained running the Triplet Finder on state-of- the-art GPU cards. In addition, a proof-of-concept system for the dispatch of data to tracking algorithms using Message Queues is presented.

GPU Particle Tracking and MHD Simulations with Greatly Enhanced Computational Speed

NASA Astrophysics Data System (ADS)

Ziemba, T.; O'Donnell, D.; Carscadden, J.; Cash, M.; Winglee, R.; Harnett, E.

2008-12-01

GPUs are intrinsically highly parallelized systems that provide more than an order of magnitude computing speed over a CPU based systems, for less cost than a high end-workstation. Recent advancements in GPU technologies allow for full IEEE float specifications with performance up to several hundred GFLOPs per GPU, and new software architectures have recently become available to ease the transition from graphics based to scientific applications. This allows for a cheap alternative to standard supercomputing methods and should increase the time to discovery. 3-D particle tracking and MHD codes have been developed using NVIDIA's CUDA and have demonstrated speed up of nearly a factor of 20 over equivalent CPU versions of the codes. Such a speed up enables new applications to develop, including real time running of radiation belt simulations and real time running of global magnetospheric simulations, both of which could provide important space weather prediction tools.

Scaling Deep Learning on GPU and Knights Landing clusters

DOE PAGES

You, Yang; Buluc, Aydin; Demmel, James

2017-09-26

The speed of deep neural networks training has become a big bottleneck of deep learning research and development. For example, training GoogleNet by ImageNet dataset on one Nvidia K20 GPU needs 21 days. To speed up the training process, the current deep learning systems heavily rely on the hardware accelerators. However, these accelerators have limited on-chip memory compared with CPUs. To handle large datasets, they need to fetch data from either CPU memory or remote processors. We use both self-hosted Intel Knights Landing (KNL) clusters and multi-GPU clusters as our target platforms. From an algorithm aspect, current distributed machine learningmore » systems are mainly designed for cloud systems. These methods are asynchronous because of the slow network and high fault-tolerance requirement on cloud systems. We focus on Elastic Averaging SGD (EASGD) to design algorithms for HPC clusters. Original EASGD used round-robin method for communication and updating. The communication is ordered by the machine rank ID, which is inefficient on HPC clusters. First, we redesign four efficient algorithms for HPC systems to improve EASGD's poor scaling on clusters. Async EASGD, Async MEASGD, and Hogwild EASGD are faster \\textcolor{black}{than} their existing counterparts (Async SGD, Async MSGD, and Hogwild SGD, resp.) in all the comparisons. Finally, we design Sync EASGD, which ties for the best performance among all the methods while being deterministic. In addition to the algorithmic improvements, we use some system-algorithm codesign techniques to scale up the algorithms. By reducing the percentage of communication from 87% to 14%, our Sync EASGD achieves 5.3x speedup over original EASGD on the same platform. We get 91.5% weak scaling efficiency on 4253 KNL cores, which is higher than the state-of-the-art implementation.« less

Scaling Deep Learning on GPU and Knights Landing clusters

DOE Office of Scientific and Technical Information (OSTI.GOV)

You, Yang; Buluc, Aydin; Demmel, James

The speed of deep neural networks training has become a big bottleneck of deep learning research and development. For example, training GoogleNet by ImageNet dataset on one Nvidia K20 GPU needs 21 days. To speed up the training process, the current deep learning systems heavily rely on the hardware accelerators. However, these accelerators have limited on-chip memory compared with CPUs. To handle large datasets, they need to fetch data from either CPU memory or remote processors. We use both self-hosted Intel Knights Landing (KNL) clusters and multi-GPU clusters as our target platforms. From an algorithm aspect, current distributed machine learningmore » systems are mainly designed for cloud systems. These methods are asynchronous because of the slow network and high fault-tolerance requirement on cloud systems. We focus on Elastic Averaging SGD (EASGD) to design algorithms for HPC clusters. Original EASGD used round-robin method for communication and updating. The communication is ordered by the machine rank ID, which is inefficient on HPC clusters. First, we redesign four efficient algorithms for HPC systems to improve EASGD's poor scaling on clusters. Async EASGD, Async MEASGD, and Hogwild EASGD are faster \\textcolor{black}{than} their existing counterparts (Async SGD, Async MSGD, and Hogwild SGD, resp.) in all the comparisons. Finally, we design Sync EASGD, which ties for the best performance among all the methods while being deterministic. In addition to the algorithmic improvements, we use some system-algorithm codesign techniques to scale up the algorithms. By reducing the percentage of communication from 87% to 14%, our Sync EASGD achieves 5.3x speedup over original EASGD on the same platform. We get 91.5% weak scaling efficiency on 4253 KNL cores, which is higher than the state-of-the-art implementation.« less

Performance evaluation for volumetric segmentation of multiple sclerosis lesions using MATLAB and computing engine in the graphical processing unit (GPU)

NASA Astrophysics Data System (ADS)

Le, Anh H.; Park, Young W.; Ma, Kevin; Jacobs, Colin; Liu, Brent J.

2010-03-01

Multiple Sclerosis (MS) is a progressive neurological disease affecting myelin pathways in the brain. Multiple lesions in the white matter can cause paralysis and severe motor disabilities of the affected patient. To solve the issue of inconsistency and user-dependency in manual lesion measurement of MRI, we have proposed a 3-D automated lesion quantification algorithm to enable objective and efficient lesion volume tracking. The computer-aided detection (CAD) of MS, written in MATLAB, utilizes K-Nearest Neighbors (KNN) method to compute the probability of lesions on a per-voxel basis. Despite the highly optimized algorithm of imaging processing that is used in CAD development, MS CAD integration and evaluation in clinical workflow is technically challenging due to the requirement of high computation rates and memory bandwidth in the recursive nature of the algorithm. In this paper, we present the development and evaluation of using a computing engine in the graphical processing unit (GPU) with MATLAB for segmentation of MS lesions. The paper investigates the utilization of a high-end GPU for parallel computing of KNN in the MATLAB environment to improve algorithm performance. The integration is accomplished using NVIDIA's CUDA developmental toolkit for MATLAB. The results of this study will validate the practicality and effectiveness of the prototype MS CAD in a clinical setting. The GPU method may allow MS CAD to rapidly integrate in an electronic patient record or any disease-centric health care system.

Comparison of a 3-D GPU-Assisted Maxwell Code and Ray Tracing for Reflectometry on ITER

NASA Astrophysics Data System (ADS)

Gady, Sarah; Kubota, Shigeyuki; Johnson, Irena

2015-11-01

Electromagnetic wave propagation and scattering in magnetized plasmas are important diagnostics for high temperature plasmas. 1-D and 2-D full-wave codes are standard tools for measurements of the electron density profile and fluctuations; however, ray tracing results have shown that beam propagation in tokamak plasmas is inherently a 3-D problem. The GPU-Assisted Maxwell Code utilizes the FDTD (Finite-Difference Time-Domain) method for solving the Maxwell equations with the cold plasma approximation in a 3-D geometry. Parallel processing with GPGPU (General-Purpose computing on Graphics Processing Units) is used to accelerate the computation. Previously, we reported on initial comparisons of the code results to 1-D numerical and analytical solutions, where the size of the computational grid was limited by the on-board memory of the GPU. In the current study, this limitation is overcome by using domain decomposition and an additional GPU. As a practical application, this code is used to study the current design of the ITER Low Field Side Reflectometer (LSFR) for the Equatorial Port Plug 11 (EPP11). A detailed examination of Gaussian beam propagation in the ITER edge plasma will be presented, as well as comparisons with ray tracing. This work was made possible by funding from the Department of Energy for the Summer Undergraduate Laboratory Internship (SULI) program. This work is supported by the US DOE Contract No.DE-AC02-09CH11466 and DE-FG02-99-ER54527.

GPU acceleration of a petascale application for turbulent mixing at high Schmidt number using OpenMP 4.5

NASA Astrophysics Data System (ADS)

Clay, M. P.; Buaria, D.; Yeung, P. K.; Gotoh, T.

2018-07-01

This paper reports on the successful implementation of a massively parallel GPU-accelerated algorithm for the direct numerical simulation of turbulent mixing at high Schmidt number. The work stems from a recent development (Comput. Phys. Commun., vol. 219, 2017, 313-328), in which a low-communication algorithm was shown to attain high degrees of scalability on the Cray XE6 architecture when overlapping communication and computation via dedicated communication threads. An even higher level of performance has now been achieved using OpenMP 4.5 on the Cray XK7 architecture, where on each node the 16 integer cores of an AMD Interlagos processor share a single Nvidia K20X GPU accelerator. In the new algorithm, data movements are minimized by performing virtually all of the intensive scalar field computations in the form of combined compact finite difference (CCD) operations on the GPUs. A memory layout in departure from usual practices is found to provide much better performance for a specific kernel required to apply the CCD scheme. Asynchronous execution enabled by adding the OpenMP 4.5 NOWAIT clause to TARGET constructs improves scalability when used to overlap computation on the GPUs with computation and communication on the CPUs. On the 27-petaflops supercomputer Titan at Oak Ridge National Laboratory, USA, a GPU-to-CPU speedup factor of approximately 5 is consistently observed at the largest problem size of 81923 grid points for the scalar field computed with 8192 XK7 nodes.

A practical implementation of 3D TTI reverse time migration with multi-GPUs

NASA Astrophysics Data System (ADS)

Li, Chun; Liu, Guofeng; Li, Yihang

2017-05-01

Tilted transversely isotropic (TTI) media are typical earth anisotropy media from practical observational studies. Accurate anisotropic imaging is recognized as a breakthrough in areas with complex anisotropic structures. TTI reverse time migration (RTM) is an important method for these areas. However, P and SV waves are coupled together in the pseudo-acoustic wave equation. The SV wave is regarded as an artifact for RTM of the P wave. We adopt matching of the anisotropy parameters to suppress the SV artifacts. Another problem in the implementation of TTI RTM is instability of the numerical solution for a variably oriented axis of symmetry. We adopt Fletcher's equation by setting a small amount of SV velocity without an acoustic approximation to stabilize the wavefield propagation. To improve calculation efficiency, we use NVIDIA graphic processing unit (GPU) with compute unified device architecture instead of traditional CPU architecture. To accomplish this, we introduced a random velocity boundary and an extended homogeneous anisotropic boundary for the remaining four anisotropic parameters in the source propagation. This process avoids large storage memory and IO requirements, which is important when using a GPU with limited bandwidth of PCI-E. Furthermore, we extend the single GPU code to multi-GPUs and present a corresponding high concurrent strategy with multiple asynchronous streams, which closely achieved an ideal speedup ratio of 2:1 when compared with a single GPU. Synthetic tests validate the correctness and effectiveness of our multi-GPUs-based TTI RTM method.

Analysis of impact of general-purpose graphics processor units in supersonic flow modeling

NASA Astrophysics Data System (ADS)

Emelyanov, V. N.; Karpenko, A. G.; Kozelkov, A. S.; Teterina, I. V.; Volkov, K. N.; Yalozo, A. V.

2017-06-01

Computational methods are widely used in prediction of complex flowfields associated with off-normal situations in aerospace engineering. Modern graphics processing units (GPU) provide architectures and new programming models that enable to harness their large processing power and to design computational fluid dynamics (CFD) simulations at both high performance and low cost. Possibilities of the use of GPUs for the simulation of external and internal flows on unstructured meshes are discussed. The finite volume method is applied to solve three-dimensional unsteady compressible Euler and Navier-Stokes equations on unstructured meshes with high resolution numerical schemes. CUDA technology is used for programming implementation of parallel computational algorithms. Solutions of some benchmark test cases on GPUs are reported, and the results computed are compared with experimental and computational data. Approaches to optimization of the CFD code related to the use of different types of memory are considered. Speedup of solution on GPUs with respect to the solution on central processor unit (CPU) is compared. Performance measurements show that numerical schemes developed achieve 20-50 speedup on GPU hardware compared to CPU reference implementation. The results obtained provide promising perspective for designing a GPU-based software framework for applications in CFD.

Parallel hyperspectral compressive sensing method on GPU

NASA Astrophysics Data System (ADS)

Bernabé, Sergio; Martín, Gabriel; Nascimento, José M. P.

2015-10-01

Remote hyperspectral sensors collect large amounts of data per flight usually with low spatial resolution. It is known that the bandwidth connection between the satellite/airborne platform and the ground station is reduced, thus a compression onboard method is desirable to reduce the amount of data to be transmitted. This paper presents a parallel implementation of an compressive sensing method, called parallel hyperspectral coded aperture (P-HYCA), for graphics processing units (GPU) using the compute unified device architecture (CUDA). This method takes into account two main properties of hyperspectral dataset, namely the high correlation existing among the spectral bands and the generally low number of endmembers needed to explain the data, which largely reduces the number of measurements necessary to correctly reconstruct the original data. Experimental results conducted using synthetic and real hyperspectral datasets on two different GPU architectures by NVIDIA: GeForce GTX 590 and GeForce GTX TITAN, reveal that the use of GPUs can provide real-time compressive sensing performance. The achieved speedup is up to 20 times when compared with the processing time of HYCA running on one core of the Intel i7-2600 CPU (3.4GHz), with 16 Gbyte memory.

Ultrafast superpixel segmentation of large 3D medical datasets

NASA Astrophysics Data System (ADS)

Leblond, Antoine; Kauffmann, Claude

2016-03-01

Even with recent hardware improvements, superpixel segmentation of large 3D medical images at interactive speed (<500 ms) remains a challenge. We will describe methods to achieve such performances using a GPU based hybrid framework implementing wavefront propagation and cellular automata resolution. Tasks will be scheduled in blocks (work units) using a wavefront propagation strategy, therefore allowing sparse scheduling. Because work units has been designed as spatially cohesive, the fast Thread Group Shared Memory can be used and reused through a Gauss-Seidel like acceleration. The work unit partitioning scheme will however vary on odd- and even-numbered iterations to reduce convergence barriers. Synchronization will be ensured by an 8-step 3D variant of the traditional Red Black Ordering scheme. An attack model and early termination will also be described and implemented as additional acceleration techniques. Using our hybrid framework and typical operating parameters, we were able to compute the superpixels of a high-resolution 512x512x512 aortic angioCT scan in 283 ms using a AMD R9 290X GPU. We achieved a 22.3X speed-up factor compared to the published reference GPU implementation.

SU (2) lattice gauge theory simulations on Fermi GPUs

NASA Astrophysics Data System (ADS)

Cardoso, Nuno; Bicudo, Pedro

2011-05-01

In this work we explore the performance of CUDA in quenched lattice SU (2) simulations. CUDA, NVIDIA Compute Unified Device Architecture, is a hardware and software architecture developed by NVIDIA for computing on the GPU. We present an analysis and performance comparison between the GPU and CPU in single and double precision. Analyses with multiple GPUs and two different architectures (G200 and Fermi architectures) are also presented. In order to obtain a high performance, the code must be optimized for the GPU architecture, i.e., an implementation that exploits the memory hierarchy of the CUDA programming model. We produce codes for the Monte Carlo generation of SU (2) lattice gauge configurations, for the mean plaquette, for the Polyakov Loop at finite T and for the Wilson loop. We also present results for the potential using many configurations (50,000) without smearing and almost 2000 configurations with APE smearing. With two Fermi GPUs we have achieved an excellent performance of 200× the speed over one CPU, in single precision, around 110 Gflops/s. We also find that, using the Fermi architecture, double precision computations for the static quark-antiquark potential are not much slower (less than 2× slower) than single precision computations.

DOE Office of Scientific and Technical Information (OSTI.GOV)

Aliaga, José I., E-mail: aliaga@uji.es; Alonso, Pedro; Badía, José M.

We introduce a new iterative Krylov subspace-based eigensolver for the simulation of macromolecular motions on desktop multithreaded platforms equipped with multicore processors and, possibly, a graphics accelerator (GPU). The method consists of two stages, with the original problem first reduced into a simpler band-structured form by means of a high-performance compute-intensive procedure. This is followed by a memory-intensive but low-cost Krylov iteration, which is off-loaded to be computed on the GPU by means of an efficient data-parallel kernel. The experimental results reveal the performance of the new eigensolver. Concretely, when applied to the simulation of macromolecules with a few thousandsmore » degrees of freedom and the number of eigenpairs to be computed is small to moderate, the new solver outperforms other methods implemented as part of high-performance numerical linear algebra packages for multithreaded architectures.« less

A high-speed DAQ framework for future high-level trigger and event building clusters

NASA Astrophysics Data System (ADS)

Caselle, M.; Ardila Perez, L. E.; Balzer, M.; Dritschler, T.; Kopmann, A.; Mohr, H.; Rota, L.; Vogelgesang, M.; Weber, M.

2017-03-01

Modern data acquisition and trigger systems require a throughput of several GB/s and latencies of the order of microseconds. To satisfy such requirements, a heterogeneous readout system based on FPGA readout cards and GPU-based computing nodes coupled by InfiniBand has been developed. The incoming data from the back-end electronics is delivered directly into the internal memory of GPUs through a dedicated peer-to-peer PCIe communication. High performance DMA engines have been developed for direct communication between FPGAs and GPUs using "DirectGMA (AMD)" and "GPUDirect (NVIDIA)" technologies. The proposed infrastructure is a candidate for future generations of event building clusters, high-level trigger filter farms and low-level trigger system. In this paper the heterogeneous FPGA-GPU architecture will be presented and its performance be discussed.

A New GPU-Enabled MODTRAN Thermal Model for the PLUME TRACKER Volcanic Emission Analysis Toolkit

NASA Astrophysics Data System (ADS)

Acharya, P. K.; Berk, A.; Guiang, C.; Kennett, R.; Perkins, T.; Realmuto, V. J.

2013-12-01

Real-time quantification of volcanic gaseous and particulate releases is important for (1) recognizing rapid increases in SO2 gaseous emissions which may signal an impending eruption; (2) characterizing ash clouds to enable safe and efficient commercial aviation; and (3) quantifying the impact of volcanic aerosols on climate forcing. The Jet Propulsion Laboratory (JPL) has developed state-of-the-art algorithms, embedded in their analyst-driven Plume Tracker toolkit, for performing SO2, NH3, and CH4 retrievals from remotely sensed multi-spectral Thermal InfraRed spectral imagery. While Plume Tracker provides accurate results, it typically requires extensive analyst time. A major bottleneck in this processing is the relatively slow but accurate FORTRAN-based MODTRAN atmospheric and plume radiance model, developed by Spectral Sciences, Inc. (SSI). To overcome this bottleneck, SSI in collaboration with JPL, is porting these slow thermal radiance algorithms onto massively parallel, relatively inexpensive and commercially-available GPUs. This paper discusses SSI's efforts to accelerate the MODTRAN thermal emission algorithms used by Plume Tracker. Specifically, we are developing a GPU implementation of the Curtis-Godson averaging and the Voigt in-band transmittances from near line center molecular absorption, which comprise the major computational bottleneck. The transmittance calculations were decomposed into separate functions, individually implemented as GPU kernels, and tested for accuracy and performance relative to the original CPU code. Speedup factors of 14 to 30× were realized for individual processing components on an NVIDIA GeForce GTX 295 graphics card with no loss of accuracy. Due to the separate host (CPU) and device (GPU) memory spaces, a redesign of the MODTRAN architecture was required to ensure efficient data transfer between host and device, and to facilitate high parallel throughput. Currently, we are incorporating the separate GPU kernels into a single function for calculating the Voigt in-band transmittance, and subsequently for integration into the re-architectured MODTRAN6 code. Our overall objective is that by combining the GPU processing with more efficient Plume Tracker retrieval algorithms, a 100-fold increase in the computational speed will be realized. Since the Plume Tracker runs on Windows-based platforms, the GPU-enhanced MODTRAN6 will be packaged as a DLL. We do however anticipate that the accelerated option will be made available to the general MODTRAN community through an application programming interface (API).

SU-G-TeP1-15: Toward a Novel GPU Accelerated Deterministic Solution to the Linear Boltzmann Transport Equation

DOE Office of Scientific and Technical Information (OSTI.GOV)

Yang, R; Fallone, B; Cross Cancer Institute, Edmonton, AB

Purpose: To develop a Graphic Processor Unit (GPU) accelerated deterministic solution to the Linear Boltzmann Transport Equation (LBTE) for accurate dose calculations in radiotherapy (RT). A deterministic solution yields the potential for major speed improvements due to the sparse matrix-vector and vector-vector multiplications and would thus be of benefit to RT. Methods: In order to leverage the massively parallel architecture of GPUs, the first order LBTE was reformulated as a second order self-adjoint equation using the Least Squares Finite Element Method (LSFEM). This produces a symmetric positive-definite matrix which is efficiently solved using a parallelized conjugate gradient (CG) solver. Themore » LSFEM formalism is applied in space, discrete ordinates is applied in angle, and the Multigroup method is applied in energy. The final linear system of equations produced is tightly coupled in space and angle. Our code written in CUDA-C was benchmarked on an Nvidia GeForce TITAN-X GPU against an Intel i7-6700K CPU. A spatial mesh of 30,950 tetrahedral elements was used with an S4 angular approximation. Results: To avoid repeating a full computationally intensive finite element matrix assembly at each Multigroup energy, a novel mapping algorithm was developed which minimized the operations required at each energy. Additionally, a parallelized memory mapping for the kronecker product between the sparse spatial and angular matrices, including Dirichlet boundary conditions, was created. Atomicity is preserved by graph-coloring overlapping nodes into separate kernel launches. The one-time mapping calculations for matrix assembly, kronecker product, and boundary condition application took 452±1ms on GPU. Matrix assembly for 16 energy groups took 556±3s on CPU, and 358±2ms on GPU using the mappings developed. The CG solver took 93±1s on CPU, and 468±2ms on GPU. Conclusion: Three computationally intensive subroutines in deterministically solving the LBTE have been formulated on GPU, resulting in two orders of magnitude speedup. Funding support from Natural Sciences and Engineering Research Council and Alberta Innovates Health Solutions. Dr. Fallone is a co-founder and CEO of MagnetTx Oncology Solutions (under discussions to license Alberta bi-planar linac MR for commercialization).« less

Protein alignment algorithms with an efficient backtracking routine on multiple GPUs.

PubMed

Blazewicz, Jacek; Frohmberg, Wojciech; Kierzynka, Michal; Pesch, Erwin; Wojciechowski, Pawel

2011-05-20

Pairwise sequence alignment methods are widely used in biological research. The increasing number of sequences is perceived as one of the upcoming challenges for sequence alignment methods in the nearest future. To overcome this challenge several GPU (Graphics Processing Unit) computing approaches have been proposed lately. These solutions show a great potential of a GPU platform but in most cases address the problem of sequence database scanning and computing only the alignment score whereas the alignment itself is omitted. Thus, the need arose to implement the global and semiglobal Needleman-Wunsch, and Smith-Waterman algorithms with a backtracking procedure which is needed to construct the alignment. In this paper we present the solution that performs the alignment of every given sequence pair, which is a required step for progressive multiple sequence alignment methods, as well as for DNA recognition at the DNA assembly stage. Performed tests show that the implementation, with performance up to 6.3 GCUPS on a single GPU for affine gap penalties, is very efficient in comparison to other CPU and GPU-based solutions. Moreover, multiple GPUs support with load balancing makes the application very scalable. The article shows that the backtracking procedure of the sequence alignment algorithms may be designed to fit in with the GPU architecture. Therefore, our algorithm, apart from scores, is able to compute pairwise alignments. This opens a wide range of new possibilities, allowing other methods from the area of molecular biology to take advantage of the new computational architecture. Performed tests show that the efficiency of the implementation is excellent. Moreover, the speed of our GPU-based algorithms can be almost linearly increased when using more than one graphics card.

Comparison and Computational Performance of Tsunami-HySEA and MOST Models for the LANTEX 2013 scenario

NASA Astrophysics Data System (ADS)

González-Vida, Jose M.; Macías, Jorge; Mercado, Aurelio; Ortega, Sergio; Castro, Manuel J.

2017-04-01

Tsunami-HySEA model is used to simulate the Caribbean LANTEX 2013 scenario (LANTEX is the acronym for Large AtlaNtic Tsunami EXercise, which is carried out annually). The numerical simulation of the propagation and inundation phases, is performed with both models but using different mesh resolutions and nested meshes. Some comparisons with the MOST tsunami model available at the University of Puerto Rico (UPR) are made. Both models compare well for propagating tsunami waves in open sea, producing very similar results. In near-shore shallow waters, Tsunami-HySEA should be compared with the inundation version of MOST, since the propagation version of MOST is limited to deeper waters. Regarding the inundation phase, a 1 arc-sec (approximately 30 m) resolution mesh covering all of Puerto Rico, is used, and a three-level nested meshes technique implemented. In the inundation phase, larger differences between model results are observed. Nevertheless, the most striking difference resides in computational time; Tsunami-HySEA is coded using the advantages of GPU architecture, and can produce a 4 h simulation in a 60 arcsec resolution grid for the whole Caribbean Sea in less than 4 min with a single general-purpose GPU and as fast as 11 s with 32 general-purpose GPUs. In the inundation stage with nested meshes, approximately 8 hours of wall clock time is needed for a 2-h simulation in a single GPU (versus more than 2 days for the MOST inundation, running three different parts of the island—West, Center, East—at the same time due to memory limitations in MOST). When domain decomposition techniques are finally implemented by breaking up the computational domain into sub-domains and assigning a GPU to each sub-domain (multi-GPU Tsunami-HySEA version), we show that the wall clock time significantly decreases, allowing high-resolution inundation modelling in very short computational times, reducing, for example, if eight GPUs are used, the wall clock time to around 1 hour. Besides, these computational times are obtained using general-purpose GPU hardware.

Livermore Big Artificial Neural Network Toolkit

DOE Office of Scientific and Technical Information (OSTI.GOV)

Essen, Brian Van; Jacobs, Sam; Kim, Hyojin

2016-07-01

LBANN is a toolkit that is designed to train artificial neural networks efficiently on high performance computing architectures. It is optimized to take advantages of key High Performance Computing features to accelerate neural network training. Specifically it is optimized for low-latency, high bandwidth interconnects, node-local NVRAM, node-local GPU accelerators, and high bandwidth parallel file systems. It is built on top of the open source Elemental distributed-memory dense and spars-direct linear algebra and optimization library that is released under the BSD license. The algorithms contained within LBANN are drawn from the academic literature and implemented to work within a distributed-memory framework.

Graphics processing unit based computation for NDE applications

NASA Astrophysics Data System (ADS)

Nahas, C. A.; Rajagopal, Prabhu; Balasubramaniam, Krishnan; Krishnamurthy, C. V.

2012-05-01

Advances in parallel processing in recent years are helping to improve the cost of numerical simulation. Breakthroughs in Graphical Processing Unit (GPU) based computation now offer the prospect of further drastic improvements. The introduction of 'compute unified device architecture' (CUDA) by NVIDIA (the global technology company based in Santa Clara, California, USA) has made programming GPUs for general purpose computing accessible to the average programmer. Here we use CUDA to develop parallel finite difference schemes as applicable to two problems of interest to NDE community, namely heat diffusion and elastic wave propagation. The implementations are for two-dimensions. Performance improvement of the GPU implementation against serial CPU implementation is then discussed.

Real-time optical flow estimation on a GPU for a skied-steered mobile robot

NASA Astrophysics Data System (ADS)

Kniaz, V. V.

2016-04-01

Accurate egomotion estimation is required for mobile robot navigation. Often the egomotion is estimated using optical flow algorithms. For an accurate estimation of optical flow most of modern algorithms require high memory resources and processor speed. However simple single-board computers that control the motion of the robot usually do not provide such resources. On the other hand, most of modern single-board computers are equipped with an embedded GPU that could be used in parallel with a CPU to improve the performance of the optical flow estimation algorithm. This paper presents a new Z-flow algorithm for efficient computation of an optical flow using an embedded GPU. The algorithm is based on the phase correlation optical flow estimation and provide a real-time performance on a low cost embedded GPU. The layered optical flow model is used. Layer segmentation is performed using graph-cut algorithm with a time derivative based energy function. Such approach makes the algorithm both fast and robust in low light and low texture conditions. The algorithm implementation for a Raspberry Pi Model B computer is discussed. For evaluation of the algorithm the computer was mounted on a Hercules mobile skied-steered robot equipped with a monocular camera. The evaluation was performed using a hardware-in-the-loop simulation and experiments with Hercules mobile robot. Also the algorithm was evaluated using KITTY Optical Flow 2015 dataset. The resulting endpoint error of the optical flow calculated with the developed algorithm was low enough for navigation of the robot along the desired trajectory.

Integrating the Nqueens Algorithm into a Parameterized Benchmark Suite

DTIC Science & Technology

2016-02-01

FOB is a 64-node heterogeneous cluster consisting of 16-IBM dx360M4 nodes, each with one NVIDIA Kepler K20M GPUs and 48-IBM dx360M4 nodes, and each...nodes have 256-GB of memory and an NVIDIA Tesla K40 GPU. More details on Excalibur can be found on the US Army DSRC website.19 Figures 3 and 4 show the

Implementing Molecular Dynamics on Hybrid High Performance Computers - Particle-Particle Particle-Mesh

DOE Office of Scientific and Technical Information (OSTI.GOV)

Brown, W Michael; Kohlmeyer, Axel; Plimpton, Steven J

The use of accelerators such as graphics processing units (GPUs) has become popular in scientific computing applications due to their low cost, impressive floating-point capabilities, high memory bandwidth, and low electrical power requirements. Hybrid high-performance computers, machines with nodes containing more than one type of floating-point processor (e.g. CPU and GPU), are now becoming more prevalent due to these advantages. In this paper, we present a continuation of previous work implementing algorithms for using accelerators into the LAMMPS molecular dynamics software for distributed memory parallel hybrid machines. In our previous work, we focused on acceleration for short-range models with anmore » approach intended to harness the processing power of both the accelerator and (multi-core) CPUs. To augment the existing implementations, we present an efficient implementation of long-range electrostatic force calculation for molecular dynamics. Specifically, we present an implementation of the particle-particle particle-mesh method based on the work by Harvey and De Fabritiis. We present benchmark results on the Keeneland InfiniBand GPU cluster. We provide a performance comparison of the same kernels compiled with both CUDA and OpenCL. We discuss limitations to parallel efficiency and future directions for improving performance on hybrid or heterogeneous computers.« less

Parallel hyperbolic PDE simulation on clusters: Cell versus GPU

NASA Astrophysics Data System (ADS)

Rostrup, Scott; De Sterck, Hans

2010-12-01

Increasingly, high-performance computing is looking towards data-parallel computational devices to enhance computational performance. Two technologies that have received significant attention are IBM's Cell Processor and NVIDIA's CUDA programming model for graphics processing unit (GPU) computing. In this paper we investigate the acceleration of parallel hyperbolic partial differential equation simulation on structured grids with explicit time integration on clusters with Cell and GPU backends. The message passing interface (MPI) is used for communication between nodes at the coarsest level of parallelism. Optimizations of the simulation code at the several finer levels of parallelism that the data-parallel devices provide are described in terms of data layout, data flow and data-parallel instructions. Optimized Cell and GPU performance are compared with reference code performance on a single x86 central processing unit (CPU) core in single and double precision. We further compare the CPU, Cell and GPU platforms on a chip-to-chip basis, and compare performance on single cluster nodes with two CPUs, two Cell processors or two GPUs in a shared memory configuration (without MPI). We finally compare performance on clusters with 32 CPUs, 32 Cell processors, and 32 GPUs using MPI. Our GPU cluster results use NVIDIA Tesla GPUs with GT200 architecture, but some preliminary results on recently introduced NVIDIA GPUs with the next-generation Fermi architecture are also included. This paper provides computational scientists and engineers who are considering porting their codes to accelerator environments with insight into how structured grid based explicit algorithms can be optimized for clusters with Cell and GPU accelerators. It also provides insight into the speed-up that may be gained on current and future accelerator architectures for this class of applications. Program summaryProgram title: SWsolver Catalogue identifier: AEGY_v1_0 Program summary URL:http://cpc.cs.qub.ac.uk/summaries/AEGY_v1_0.html Program obtainable from: CPC Program Library, Queen's University, Belfast, N. Ireland Licensing provisions: GPL v3 No. of lines in distributed program, including test data, etc.: 59 168 No. of bytes in distributed program, including test data, etc.: 453 409 Distribution format: tar.gz Programming language: C, CUDA Computer: Parallel Computing Clusters. Individual compute nodes may consist of x86 CPU, Cell processor, or x86 CPU with attached NVIDIA GPU accelerator. Operating system: Linux Has the code been vectorised or parallelized?: Yes. Tested on 1-128 x86 CPU cores, 1-32 Cell Processors, and 1-32 NVIDIA GPUs. RAM: Tested on Problems requiring up to 4 GB per compute node. Classification: 12 External routines: MPI, CUDA, IBM Cell SDK Nature of problem: MPI-parallel simulation of Shallow Water equations using high-resolution 2D hyperbolic equation solver on regular Cartesian grids for x86 CPU, Cell Processor, and NVIDIA GPU using CUDA. Solution method: SWsolver provides 3 implementations of a high-resolution 2D Shallow Water equation solver on regular Cartesian grids, for CPU, Cell Processor, and NVIDIA GPU. Each implementation uses MPI to divide work across a parallel computing cluster. Additional comments: Sub-program numdiff is used for the test run.

Acceleration of the Smith-Waterman algorithm using single and multiple graphics processors

NASA Astrophysics Data System (ADS)

Khajeh-Saeed, Ali; Poole, Stephen; Blair Perot, J.

2010-06-01

Finding regions of similarity between two very long data streams is a computationally intensive problem referred to as sequence alignment. Alignment algorithms must allow for imperfect sequence matching with different starting locations and some gaps and errors between the two data sequences. Perhaps the most well known application of sequence matching is the testing of DNA or protein sequences against genome databases. The Smith-Waterman algorithm is a method for precisely characterizing how well two sequences can be aligned and for determining the optimal alignment of those two sequences. Like many applications in computational science, the Smith-Waterman algorithm is constrained by the memory access speed and can be accelerated significantly by using graphics processors (GPUs) as the compute engine. In this work we show that effective use of the GPU requires a novel reformulation of the Smith-Waterman algorithm. The performance of this new version of the algorithm is demonstrated using the SSCA#1 (Bioinformatics) benchmark running on one GPU and on up to four GPUs executing in parallel. The results indicate that for large problems a single GPU is up to 45 times faster than a CPU for this application, and the parallel implementation shows linear speed up on up to 4 GPUs.

Efficient Irregular Wavefront Propagation Algorithms on Hybrid CPU-GPU Machines

PubMed Central

Teodoro, George; Pan, Tony; Kurc, Tahsin; Kong, Jun; Cooper, Lee; Saltz, Joel

2013-01-01

We address the problem of efficient execution of a computation pattern, referred to here as the irregular wavefront propagation pattern (IWPP), on hybrid systems with multiple CPUs and GPUs. The IWPP is common in several image processing operations. In the IWPP, data elements in the wavefront propagate waves to their neighboring elements on a grid if a propagation condition is satisfied. Elements receiving the propagated waves become part of the wavefront. This pattern results in irregular data accesses and computations. We develop and evaluate strategies for efficient computation and propagation of wavefronts using a multi-level queue structure. This queue structure improves the utilization of fast memories in a GPU and reduces synchronization overheads. We also develop a tile-based parallelization strategy to support execution on multiple CPUs and GPUs. We evaluate our approaches on a state-of-the-art GPU accelerated machine (equipped with 3 GPUs and 2 multicore CPUs) using the IWPP implementations of two widely used image processing operations: morphological reconstruction and euclidean distance transform. Our results show significant performance improvements on GPUs. The use of multiple CPUs and GPUs cooperatively attains speedups of 50× and 85× with respect to single core CPU executions for morphological reconstruction and euclidean distance transform, respectively. PMID:23908562

Porting marine ecosystem model spin-up using transport matrices to GPUs

NASA Astrophysics Data System (ADS)

Siewertsen, E.; Piwonski, J.; Slawig, T.

2013-01-01

We have ported an implementation of the spin-up for marine ecosystem models based on transport matrices to graphics processing units (GPUs). The original implementation was designed for distributed-memory architectures and uses the Portable, Extensible Toolkit for Scientific Computation (PETSc) library that is based on the Message Passing Interface (MPI) standard. The spin-up computes a steady seasonal cycle of ecosystem tracers with climatological ocean circulation data as forcing. Since the transport is linear with respect to the tracers, the resulting operator is represented by matrices. Each iteration of the spin-up involves two matrix-vector multiplications and the evaluation of the used biogeochemical model. The original code was written in C and Fortran. On the GPU, we use the Compute Unified Device Architecture (CUDA) standard, a customized version of PETSc and a commercial CUDA Fortran compiler. We describe the extensions to PETSc and the modifications of the original C and Fortran codes that had to be done. Here we make use of freely available libraries for the GPU. We analyze the computational effort of the main parts of the spin-up for two exemplar ecosystem models and compare the overall computational time to those necessary on different CPUs. The results show that a consumer GPU can compete with a significant number of cluster CPUs without further code optimization.

SU (2) lattice gauge theory simulations on Fermi GPUs

DOE Office of Scientific and Technical Information (OSTI.GOV)

Cardoso, Nuno, E-mail: nunocardoso@cftp.ist.utl.p; Bicudo, Pedro, E-mail: bicudo@ist.utl.p

2011-05-10

In this work we explore the performance of CUDA in quenched lattice SU (2) simulations. CUDA, NVIDIA Compute Unified Device Architecture, is a hardware and software architecture developed by NVIDIA for computing on the GPU. We present an analysis and performance comparison between the GPU and CPU in single and double precision. Analyses with multiple GPUs and two different architectures (G200 and Fermi architectures) are also presented. In order to obtain a high performance, the code must be optimized for the GPU architecture, i.e., an implementation that exploits the memory hierarchy of the CUDA programming model. We produce codes formore » the Monte Carlo generation of SU (2) lattice gauge configurations, for the mean plaquette, for the Polyakov Loop at finite T and for the Wilson loop. We also present results for the potential using many configurations (50,000) without smearing and almost 2000 configurations with APE smearing. With two Fermi GPUs we have achieved an excellent performance of 200x the speed over one CPU, in single precision, around 110 Gflops/s. We also find that, using the Fermi architecture, double precision computations for the static quark-antiquark potential are not much slower (less than 2x slower) than single precision computations.« less

GRAVIDY, a GPU modular, parallel direct-summation N-body integrator: dynamics with softening

NASA Astrophysics Data System (ADS)

Maureira-Fredes, Cristián; Amaro-Seoane, Pau

2018-01-01

A wide variety of outstanding problems in astrophysics involve the motion of a large number of particles under the force of gravity. These include the global evolution of globular clusters, tidal disruptions of stars by a massive black hole, the formation of protoplanets and sources of gravitational radiation. The direct-summation of N gravitational forces is a complex problem with no analytical solution and can only be tackled with approximations and numerical methods. To this end, the Hermite scheme is a widely used integration method. With different numerical techniques and special-purpose hardware, it can be used to speed up the calculations. But these methods tend to be computationally slow and cumbersome to work with. We present a new graphics processing unit (GPU), direct-summation N-body integrator written from scratch and based on this scheme, which includes relativistic corrections for sources of gravitational radiation. GRAVIDY has high modularity, allowing users to readily introduce new physics, it exploits available computational resources and will be maintained by regular updates. GRAVIDY can be used in parallel on multiple CPUs and GPUs, with a considerable speed-up benefit. The single-GPU version is between one and two orders of magnitude faster than the single-CPU version. A test run using four GPUs in parallel shows a speed-up factor of about 3 as compared to the single-GPU version. The conception and design of this first release is aimed at users with access to traditional parallel CPU clusters or computational nodes with one or a few GPU cards.

GPU-Accelerated Molecular Dynamics Simulation to Study Liquid Crystal Phase Transition Using Coarse-Grained Gay-Berne Anisotropic Potential.

PubMed

Chen, Wenduo; Zhu, Youliang; Cui, Fengchao; Liu, Lunyang; Sun, Zhaoyan; Chen, Jizhong; Li, Yunqi

2016-01-01

Gay-Berne (GB) potential is regarded as an accurate model in the simulation of anisotropic particles, especially for liquid crystal (LC) mesogens. However, its computational complexity leads to an extremely time-consuming process for large systems. Here, we developed a GPU-accelerated molecular dynamics (MD) simulation with coarse-grained GB potential implemented in GALAMOST package to investigate the LC phase transitions for mesogens in small molecules, main-chain or side-chain polymers. For identical mesogens in three different molecules, on cooling from fully isotropic melts, the small molecules form a single-domain smectic-B phase, while the main-chain LC polymers prefer a single-domain nematic phase as a result of connective restraints in neighboring mesogens. The phase transition of side-chain LC polymers undergoes a two-step process: nucleation of nematic islands and formation of multi-domain nematic texture. The particular behavior originates in the fact that the rotational orientation of the mesogenes is hindered by the polymer backbones. Both the global distribution and the local orientation of mesogens are critical for the phase transition of anisotropic particles. Furthermore, compared with the MD simulation in LAMMPS, our GPU-accelerated code is about 4 times faster than the GPU version of LAMMPS and at least 200 times faster than the CPU version of LAMMPS. This study clearly shows that GPU-accelerated MD simulation with GB potential in GALAMOST can efficiently handle systems with anisotropic particles and interactions, and accurately explore phase differences originated from molecular structures.

GPU-Accelerated Molecular Dynamics Simulation to Study Liquid Crystal Phase Transition Using Coarse-Grained Gay-Berne Anisotropic Potential

PubMed Central

Cui, Fengchao; Liu, Lunyang; Sun, Zhaoyan; Chen, Jizhong; Li, Yunqi

2016-01-01

Gay-Berne (GB) potential is regarded as an accurate model in the simulation of anisotropic particles, especially for liquid crystal (LC) mesogens. However, its computational complexity leads to an extremely time-consuming process for large systems. Here, we developed a GPU-accelerated molecular dynamics (MD) simulation with coarse-grained GB potential implemented in GALAMOST package to investigate the LC phase transitions for mesogens in small molecules, main-chain or side-chain polymers. For identical mesogens in three different molecules, on cooling from fully isotropic melts, the small molecules form a single-domain smectic-B phase, while the main-chain LC polymers prefer a single-domain nematic phase as a result of connective restraints in neighboring mesogens. The phase transition of side-chain LC polymers undergoes a two-step process: nucleation of nematic islands and formation of multi-domain nematic texture. The particular behavior originates in the fact that the rotational orientation of the mesogenes is hindered by the polymer backbones. Both the global distribution and the local orientation of mesogens are critical for the phase transition of anisotropic particles. Furthermore, compared with the MD simulation in LAMMPS, our GPU-accelerated code is about 4 times faster than the GPU version of LAMMPS and at least 200 times faster than the CPU version of LAMMPS. This study clearly shows that GPU-accelerated MD simulation with GB potential in GALAMOST can efficiently handle systems with anisotropic particles and interactions, and accurately explore phase differences originated from molecular structures. PMID:26986851

Parallel k-means++ for Multiple Shared-Memory Architectures

DOE Office of Scientific and Technical Information (OSTI.GOV)

Mackey, Patrick S.; Lewis, Robert R.

2016-09-22

In recent years k-means++ has become a popular initialization technique for improved k-means clustering. To date, most of the work done to improve its performance has involved parallelizing algorithms that are only approximations of k-means++. In this paper we present a parallelization of the exact k-means++ algorithm, with a proof of its correctness. We develop implementations for three distinct shared-memory architectures: multicore CPU, high performance GPU, and the massively multithreaded Cray XMT platform. We demonstrate the scalability of the algorithm on each platform. In addition we present a visual approach for showing which platform performed k-means++ the fastest for varyingmore » data sizes.« less

cuTauLeaping: A GPU-Powered Tau-Leaping Stochastic Simulator for Massive Parallel Analyses of Biological Systems

PubMed Central

Besozzi, Daniela; Pescini, Dario; Mauri, Giancarlo

2014-01-01

Tau-leaping is a stochastic simulation algorithm that efficiently reconstructs the temporal evolution of biological systems, modeled according to the stochastic formulation of chemical kinetics. The analysis of dynamical properties of these systems in physiological and perturbed conditions usually requires the execution of a large number of simulations, leading to high computational costs. Since each simulation can be executed independently from the others, a massive parallelization of tau-leaping can bring to relevant reductions of the overall running time. The emerging field of General Purpose Graphic Processing Units (GPGPU) provides power-efficient high-performance computing at a relatively low cost. In this work we introduce cuTauLeaping, a stochastic simulator of biological systems that makes use of GPGPU computing to execute multiple parallel tau-leaping simulations, by fully exploiting the Nvidia's Fermi GPU architecture. We show how a considerable computational speedup is achieved on GPU by partitioning the execution of tau-leaping into multiple separated phases, and we describe how to avoid some implementation pitfalls related to the scarcity of memory resources on the GPU streaming multiprocessors. Our results show that cuTauLeaping largely outperforms the CPU-based tau-leaping implementation when the number of parallel simulations increases, with a break-even directly depending on the size of the biological system and on the complexity of its emergent dynamics. In particular, cuTauLeaping is exploited to investigate the probability distribution of bistable states in the Schlögl model, and to carry out a bidimensional parameter sweep analysis to study the oscillatory regimes in the Ras/cAMP/PKA pathway in S. cerevisiae. PMID:24663957

Bin-Hash Indexing: A Parallel Method for Fast Query Processing

DOE Office of Scientific and Technical Information (OSTI.GOV)

Bethel, Edward W; Gosink, Luke J.; Wu, Kesheng

2008-06-27

This paper presents a new parallel indexing data structure for answering queries. The index, called Bin-Hash, offers extremely high levels of concurrency, and is therefore well-suited for the emerging commodity of parallel processors, such as multi-cores, cell processors, and general purpose graphics processing units (GPU). The Bin-Hash approach first bins the base data, and then partitions and separately stores the values in each bin as a perfect spatial hash table. To answer a query, we first determine whether or not a record satisfies the query conditions based on the bin boundaries. For the bins with records that can not bemore » resolved, we examine the spatial hash tables. The procedures for examining the bin numbers and the spatial hash tables offer the maximum possible level of concurrency; all records are able to be evaluated by our procedure independently in parallel. Additionally, our Bin-Hash procedures access much smaller amounts of data than similar parallel methods, such as the projection index. This smaller data footprint is critical for certain parallel processors, like GPUs, where memory resources are limited. To demonstrate the effectiveness of Bin-Hash, we implement it on a GPU using the data-parallel programming language CUDA. The concurrency offered by the Bin-Hash index allows us to fully utilize the GPU's massive parallelism in our work; over 12,000 records can be simultaneously evaluated at any one time. We show that our new query processing method is an order of magnitude faster than current state-of-the-art CPU-based indexing technologies. Additionally, we compare our performance to existing GPU-based projection index strategies.« less

A multi-GPU real-time dose simulation software framework for lung radiotherapy.

PubMed

Santhanam, A P; Min, Y; Neelakkantan, H; Papp, N; Meeks, S L; Kupelian, P A

2012-09-01

Medical simulation frameworks facilitate both the preoperative and postoperative analysis of the patient's pathophysical condition. Of particular importance is the simulation of radiation dose delivery for real-time radiotherapy monitoring and retrospective analyses of the patient's treatment. In this paper, a software framework tailored for the development of simulation-based real-time radiation dose monitoring medical applications is discussed. A multi-GPU-based computational framework coupled with inter-process communication methods is introduced for simulating the radiation dose delivery on a deformable 3D volumetric lung model and its real-time visualization. The model deformation and the corresponding dose calculation are allocated among the GPUs in a task-specific manner and is performed in a pipelined manner. Radiation dose calculations are computed on two different GPU hardware architectures. The integration of this computational framework with a front-end software layer and back-end patient database repository is also discussed. Real-time simulation of the dose delivered is achieved at once every 120 ms using the proposed framework. With a linear increase in the number of GPU cores, the computational time of the simulation was linearly decreased. The inter-process communication time also improved with an increase in the hardware memory. Variations in the delivered dose and computational speedup for variations in the data dimensions are investigated using D70 and D90 as well as gEUD as metrics for a set of 14 patients. Computational speed-up increased with an increase in the beam dimensions when compared with a CPU-based commercial software while the error in the dose calculation was <1%. Our analyses show that the framework applied to deformable lung model-based radiotherapy is an effective tool for performing both real-time and retrospective analyses.

Opticks : GPU Optical Photon Simulation for Particle Physics using NVIDIA® OptiX™

NASA Astrophysics Data System (ADS)

C, Blyth Simon

2017-10-01

Opticks is an open source project that integrates the NVIDIA OptiX GPU ray tracing engine with Geant4 toolkit based simulations. Massive parallelism brings drastic performance improvements with optical photon simulation speedup expected to exceed 1000 times Geant4 when using workstation GPUs. Optical photon simulation time becomes effectively zero compared to the rest of the simulation. Optical photons from scintillation and Cherenkov processes are allocated, generated and propagated entirely on the GPU, minimizing transfer overheads and allowing CPU memory usage to be restricted to optical photons that hit photomultiplier tubes or other photon detectors. Collecting hits into standard Geant4 hit collections then allows the rest of the simulation chain to proceed unmodified. Optical physics processes of scattering, absorption, scintillator reemission and boundary processes are implemented in CUDA OptiX programs based on the Geant4 implementations. Wavelength dependent material and surface properties as well as inverse cumulative distribution functions for reemission are interleaved into GPU textures providing fast interpolated property lookup or wavelength generation. Geometry is provided to OptiX in the form of CUDA programs that return bounding boxes for each primitive and ray geometry intersection positions. Some critical parts of the geometry such as photomultiplier tubes have been implemented analytically with the remainder being tessellated. OptiX handles the creation and application of a choice of acceleration structures such as boundary volume hierarchies and the transparent use of multiple GPUs. OptiX supports interoperation with OpenGL and CUDA Thrust that has enabled unprecedented visualisations of photon propagations to be developed using OpenGL geometry shaders to provide interactive time scrubbing and CUDA Thrust photon indexing to enable interactive history selection.

GPU-based Branchless Distance-Driven Projection and Backprojection

PubMed Central

Liu, Rui; Fu, Lin; De Man, Bruno; Yu, Hengyong

2017-01-01

Projection and backprojection operations are essential in a variety of image reconstruction and physical correction algorithms in CT. The distance-driven (DD) projection and backprojection are widely used for their highly sequential memory access pattern and low arithmetic cost. However, a typical DD implementation has an inner loop that adjusts the calculation depending on the relative position between voxel and detector cell boundaries. The irregularity of the branch behavior makes it inefficient to be implemented on massively parallel computing devices such as graphics processing units (GPUs). Such irregular branch behaviors can be eliminated by factorizing the DD operation as three branchless steps: integration, linear interpolation, and differentiation, all of which are highly amenable to massive vectorization. In this paper, we implement and evaluate a highly parallel branchless DD algorithm for 3D cone beam CT. The algorithm utilizes the texture memory and hardware interpolation on GPUs to achieve fast computational speed. The developed branchless DD algorithm achieved 137-fold speedup for forward projection and 188-fold speedup for backprojection relative to a single-thread CPU implementation. Compared with a state-of-the-art 32-thread CPU implementation, the proposed branchless DD achieved 8-fold acceleration for forward projection and 10-fold acceleration for backprojection. GPU based branchless DD method was evaluated by iterative reconstruction algorithms with both simulation and real datasets. It obtained visually identical images as the CPU reference algorithm. PMID:29333480

GPU-based Branchless Distance-Driven Projection and Backprojection.

PubMed

Liu, Rui; Fu, Lin; De Man, Bruno; Yu, Hengyong

2017-12-01

Projection and backprojection operations are essential in a variety of image reconstruction and physical correction algorithms in CT. The distance-driven (DD) projection and backprojection are widely used for their highly sequential memory access pattern and low arithmetic cost. However, a typical DD implementation has an inner loop that adjusts the calculation depending on the relative position between voxel and detector cell boundaries. The irregularity of the branch behavior makes it inefficient to be implemented on massively parallel computing devices such as graphics processing units (GPUs). Such irregular branch behaviors can be eliminated by factorizing the DD operation as three branchless steps: integration, linear interpolation, and differentiation, all of which are highly amenable to massive vectorization. In this paper, we implement and evaluate a highly parallel branchless DD algorithm for 3D cone beam CT. The algorithm utilizes the texture memory and hardware interpolation on GPUs to achieve fast computational speed. The developed branchless DD algorithm achieved 137-fold speedup for forward projection and 188-fold speedup for backprojection relative to a single-thread CPU implementation. Compared with a state-of-the-art 32-thread CPU implementation, the proposed branchless DD achieved 8-fold acceleration for forward projection and 10-fold acceleration for backprojection. GPU based branchless DD method was evaluated by iterative reconstruction algorithms with both simulation and real datasets. It obtained visually identical images as the CPU reference algorithm.

Accelerating 3D Hall MHD Magnetosphere Simulations with Graphics Processing Units

NASA Astrophysics Data System (ADS)

Bard, C.; Dorelli, J.

2017-12-01

The resolution required to simulate planetary magnetospheres with Hall magnetohydrodynamics result in program sizes approaching several hundred million grid cells. These would take years to run on a single computational core and require hundreds or thousands of computational cores to complete in a reasonable time. However, this requires access to the largest supercomputers. Graphics processing units (GPUs) provide a viable alternative: one GPU can do the work of roughly 100 cores, bringing Hall MHD simulations of Ganymede within reach of modest GPU clusters ( 8 GPUs). We report our progress in developing a GPU-accelerated, three-dimensional Hall magnetohydrodynamic code and present Hall MHD simulation results for both Ganymede (run on 8 GPUs) and Mercury (56 GPUs). We benchmark our Ganymede simulation with previous results for the Galileo G8 flyby, namely that adding the Hall term to ideal MHD simulations changes the global convection pattern within the magnetosphere. Additionally, we present new results for the G1 flyby as well as initial results from Hall MHD simulations of Mercury and compare them with the corresponding ideal MHD runs.

Lattice Boltzmann multi-phase simulations in porous media using Multiple GPUs

NASA Astrophysics Data System (ADS)

Toelke, J.; De Prisco, G.; Mu, Y.

2011-12-01

Ingrain's digital rock physics lab computes the physical properties and fluid flow characteristics of oil and gas reservoir rocks including shales, carbonates and sandstones. Ingrain uses advanced lattice Boltzmann methods (LBM) to simulate multiphase flow in the rocks (porous media). We present a very efficient implementation of these methods based on CUDA. Because LBM operates on a finite difference grid, is explicit in nature, and requires only next-neighbor interactions, it is suitable for implementation on GPUs. Since GPU hardware allows for very fine grain parallelism, every lattice site can be handled by a different core. Data has to be loaded from and stored to the device memory in such a way that dense access to the memory is ensured. This can be achieved by accessing the lattice nodes with respect to their contiguous memory locations [1,2]. The simulation engine uses a sparse data structure to represent the grid and advanced algorithms to handle the moving fluid-fluid interface. The simulations are accelerated on one GPU by one order of magnitude compared to a state of the art multicore desktop computer. The engine is parallelized using MPI and runs on multiple GPUs in the same node or across the Infiniband network. Simulations with up to 50 GPUs in parallel are presented. With this simulator using it is possible to perform pore scale multi-phase (oil-water-matrix) simulations in natural porous media in a commercial manner and to predict important rock properties like absolute permeability, relative permeabilites and capillary pressure [3,4]. Results and videos of these simulations in complex real world porous media and rocks are presented and discussed.

Memory transfer optimization for a lattice Boltzmann solver on Kepler architecture nVidia GPUs

NASA Astrophysics Data System (ADS)

Mawson, Mark J.; Revell, Alistair J.

2014-10-01

The Lattice Boltzmann method (LBM) for solving fluid flow is naturally well suited to an efficient implementation for massively parallel computing, due to the prevalence of local operations in the algorithm. This paper presents and analyses the performance of a 3D lattice Boltzmann solver, optimized for third generation nVidia GPU hardware, also known as 'Kepler'. We provide a review of previous optimization strategies and analyse data read/write times for different memory types. In LBM, the time propagation step (known as streaming), involves shifting data to adjacent locations and is central to parallel performance; here we examine three approaches which make use of different hardware options. Two of which make use of 'performance enhancing' features of the GPU; shared memory and the new shuffle instruction found in Kepler based GPUs. These are compared to a standard transfer of data which relies instead on optimized storage to increase coalesced access. It is shown that the more simple approach is most efficient; since the need for large numbers of registers per thread in LBM limits the block size and thus the efficiency of these special features is reduced. Detailed results are obtained for a D3Q19 LBM solver, which is benchmarked on nVidia K5000M and K20C GPUs. In the latter case the use of a read-only data cache is explored, and peak performance of over 1036 Million Lattice Updates Per Second (MLUPS) is achieved. The appearance of a periodic bottleneck in the solver performance is also reported, believed to be hardware related; spikes in iteration-time occur with a frequency of around 11 Hz for both GPUs, independent of the size of the problem.

Efficient gaussian density formulation of volume and surface areas of macromolecules on graphical processing units.

PubMed

Zhang, Baofeng; Kilburg, Denise; Eastman, Peter; Pande, Vijay S; Gallicchio, Emilio

2017-04-15

We present an algorithm to efficiently compute accurate volumes and surface areas of macromolecules on graphical processing unit (GPU) devices using an analytic model which represents atomic volumes by continuous Gaussian densities. The volume of the molecule is expressed by means of the inclusion-exclusion formula, which is based on the summation of overlap integrals among multiple atomic densities. The surface area of the molecule is obtained by differentiation of the molecular volume with respect to atomic radii. The many-body nature of the model makes a port to GPU devices challenging. To our knowledge, this is the first reported full implementation of this model on GPU hardware. To accomplish this, we have used recursive strategies to construct the tree of overlaps and to accumulate volumes and their gradients on the tree data structures so as to minimize memory contention. The algorithm is used in the formulation of a surface area-based non-polar implicit solvent model implemented as an open source plug-in (named GaussVol) for the popular OpenMM library for molecular mechanics modeling. GaussVol is 50 to 100 times faster than our best optimized implementation for the CPUs, achieving speeds in excess of 100 ns/day with 1 fs time-step for protein-sized systems on commodity GPUs. © 2017 Wiley Periodicals, Inc. © 2017 Wiley Periodicals, Inc.

Application of a GPU-Assisted Maxwell Code to Electromagnetic Wave Propagation in ITER

NASA Astrophysics Data System (ADS)

Kubota, S.; Peebles, W. A.; Woodbury, D.; Johnson, I.; Zolfaghari, A.

2014-10-01

The Low Field Side Reflectometer (LSFR) on ITER is envisioned to provide capabilities for electron density profile and fluctuations measurements in both the plasma core and edge. The current design for the Equatorial Port Plug 11 (EPP11) employs seven monostatic antennas for use with both fixed-frequency and swept-frequency systems. The present work examines the characteristics of this layout using the 3-D version of the GPU-Assisted Maxwell Code (GAMC-3D). Previous studies in this area were performed with either 2-D full wave codes or 3-D ray- and beam-tracing. GAMC-3D is based on the FDTD method and can be run with either a fixed-frequency or modulated (e.g. FMCW) source, and with either a stationary or moving target (e.g. Doppler backscattering). The code is designed to run on a single NVIDIA Tesla GPU accelerator, and utilizes a technique based on the moving window method to overcome the size limitation of the onboard memory. Effects such as beam drift, linear mode conversion, and diffraction/scattering will be examined. Comparisons will be made with beam-tracing calculations using the complex eikonal method. Supported by U.S. DoE Grants DE-FG02-99ER54527 and DE-AC02-09CH11466, and the DoE SULI Program at PPPL.

Latest generation interconnect technologies in APEnet+ networking infrastructure

NASA Astrophysics Data System (ADS)

Ammendola, Roberto; Biagioni, Andrea; Cretaro, Paolo; Frezza, Ottorino; Lo Cicero, Francesca; Lonardo, Alessandro; Martinelli, Michele; Stanislao Paolucci, Pier; Pastorelli, Elena; Rossetti, Davide; Simula, Francesco; Vicini, Piero

2017-10-01

In this paper we present the status of the 3rd generation design of the APEnet board (V5) built upon the 28nm Altera Stratix V FPGA; it features a PCIe Gen3 x8 interface and enhanced embedded transceivers with a maximum capability of 12.5Gbps each. The network architecture is designed in accordance to the Remote DMA paradigm. The APEnet+ V5 prototype is built upon the Stratix V DevKit with the addition of a proprietary, third party IP core implementing multi-DMA engines. Support for zero-copy communication is assured by the possibility of DMA-accessing either host and GPU memory, offloading the CPU from the chore of data copying. The current implementation plateaus to a bandwidth for memory read of 4.8GB/s. Here we describe the hardware optimization to the memory write process which relies on the use of two independent DMA engines and an improved TLB.

Extending the length and time scales of Gram-Schmidt Lyapunov vector computations

NASA Astrophysics Data System (ADS)

Costa, Anthony B.; Green, Jason R.

2013-08-01

Lyapunov vectors have found growing interest recently due to their ability to characterize systems out of thermodynamic equilibrium. The computation of orthogonal Gram-Schmidt vectors requires multiplication and QR decomposition of large matrices, which grow as N2 (with the particle count). This expense has limited such calculations to relatively small systems and short time scales. Here, we detail two implementations of an algorithm for computing Gram-Schmidt vectors. The first is a distributed-memory message-passing method using Scalapack. The second uses the newly-released MAGMA library for GPUs. We compare the performance of both codes for Lennard-Jones fluids from N=100 to 1300 between Intel Nahalem/Infiniband DDR and NVIDIA C2050 architectures. To our best knowledge, these are the largest systems for which the Gram-Schmidt Lyapunov vectors have been computed, and the first time their calculation has been GPU-accelerated. We conclude that Lyapunov vector calculations can be significantly extended in length and time by leveraging the power of GPU-accelerated linear algebra.

An Analysis of Scalable GPU-Based Ray-Guided Volume Rendering

PubMed Central

Fogal, Thomas; Schiewe, Alexander; Krüger, Jens

2014-01-01

Volume rendering continues to be a critical method for analyzing large-scale scalar fields, in disciplines as diverse as biomedical engineering and computational fluid dynamics. Commodity desktop hardware has struggled to keep pace with data size increases, challenging modern visualization software to deliver responsive interactions for O(N3) algorithms such as volume rendering. We target the data type common in these domains: regularly-structured data. In this work, we demonstrate that the major limitation of most volume rendering approaches is their inability to switch the data sampling rate (and thus data size) quickly. Using a volume renderer inspired by recent work, we demonstrate that the actual amount of visualizable data for a scene is typically bound considerably lower than the memory available on a commodity GPU. Our instrumented renderer is used to investigate design decisions typically swept under the rug in volume rendering literature. The renderer is freely available, with binaries for all major platforms as well as full source code, to encourage reproduction and comparison with future research. PMID:25506079

Graphical processors for HEP trigger systems

NASA Astrophysics Data System (ADS)

Ammendola, R.; Biagioni, A.; Chiozzi, S.; Cotta Ramusino, A.; Di Lorenzo, S.; Fantechi, R.; Fiorini, M.; Frezza, O.; Lamanna, G.; Lo Cicero, F.; Lonardo, A.; Martinelli, M.; Neri, I.; Paolucci, P. S.; Pastorelli, E.; Piandani, R.; Pontisso, L.; Rossetti, D.; Simula, F.; Sozzi, M.; Vicini, P.

2017-02-01

General-purpose computing on GPUs is emerging as a new paradigm in several fields of science, although so far applications have been tailored to employ GPUs as accelerators in offline computations. With the steady decrease of GPU latencies and the increase in link and memory throughputs, time is ripe for real-time applications using GPUs in high-energy physics data acquisition and trigger systems. We will discuss the use of online parallel computing on GPUs for synchronous low level trigger systems, focusing on tests performed on the trigger of the CERN NA62 experiment. Latencies of all components need analysing, networking being the most critical. To keep it under control, we envisioned NaNet, an FPGA-based PCIe Network Interface Card (NIC) enabling GPUDirect connection. Moreover, we discuss how specific trigger algorithms can be parallelised and thus benefit from a GPU implementation, in terms of increased execution speed. Such improvements are particularly relevant for the foreseen LHC luminosity upgrade where highly selective algorithms will be crucial to maintain sustainable trigger rates with very high pileup.

Real time emotion aware applications: a case study employing emotion evocative pictures and neuro-physiological sensing enhanced by Graphic Processor Units.

PubMed

Konstantinidis, Evdokimos I; Frantzidis, Christos A; Pappas, Costas; Bamidis, Panagiotis D

2012-07-01

In this paper the feasibility of adopting Graphic Processor Units towards real-time emotion aware computing is investigated for boosting the time consuming computations employed in such applications. The proposed methodology was employed in analysis of encephalographic and electrodermal data gathered when participants passively viewed emotional evocative stimuli. The GPU effectiveness when processing electroencephalographic and electrodermal recordings is demonstrated by comparing the execution time of chaos/complexity analysis through nonlinear dynamics (multi-channel correlation dimension/D2) and signal processing algorithms (computation of skin conductance level/SCL) into various popular programming environments. Apart from the beneficial role of parallel programming, the adoption of special design techniques regarding memory management may further enhance the time minimization which approximates a factor of 30 in comparison with ANSI C language (single-core sequential execution). Therefore, the use of GPU parallel capabilities offers a reliable and robust solution for real-time sensing the user's affective state. Copyright © 2012 Elsevier Ireland Ltd. All rights reserved.

FPGA Implementation of the Coupled Filtering Method and the Affine Warping Method.

PubMed

Zhang, Chen; Liang, Tianzhu; Mok, Philip K T; Yu, Weichuan

2017-07-01

In ultrasound image analysis, the speckle tracking methods are widely applied to study the elasticity of body tissue. However, "feature-motion decorrelation" still remains as a challenge for the speckle tracking methods. Recently, a coupled filtering method and an affine warping method were proposed to accurately estimate strain values, when the tissue deformation is large. The major drawback of these methods is the high computational complexity. Even the graphics processing unit (GPU)-based program requires a long time to finish the analysis. In this paper, we propose field-programmable gate array (FPGA)-based implementations of both methods for further acceleration. The capability of FPGAs on handling different image processing components in these methods is discussed. A fast and memory-saving image warping approach is proposed. The algorithms are reformulated to build a highly efficient pipeline on FPGA. The final implementations on a Xilinx Virtex-7 FPGA are at least 13 times faster than the GPU implementation on the NVIDIA graphic card (GeForce GTX 580).

Accelerating epistasis analysis in human genetics with consumer graphics hardware.

PubMed

Sinnott-Armstrong, Nicholas A; Greene, Casey S; Cancare, Fabio; Moore, Jason H

2009-07-24

Human geneticists are now capable of measuring more than one million DNA sequence variations from across the human genome. The new challenge is to develop computationally feasible methods capable of analyzing these data for associations with common human disease, particularly in the context of epistasis. Epistasis describes the situation where multiple genes interact in a complex non-linear manner to determine an individual's disease risk and is thought to be ubiquitous for common diseases. Multifactor Dimensionality Reduction (MDR) is an algorithm capable of detecting epistasis. An exhaustive analysis with MDR is often computationally expensive, particularly for high order interactions. This challenge has previously been met with parallel computation and expensive hardware. The option we examine here exploits commodity hardware designed for computer graphics. In modern computers Graphics Processing Units (GPUs) have more memory bandwidth and computational capability than Central Processing Units (CPUs) and are well suited to this problem. Advances in the video game industry have led to an economy of scale creating a situation where these powerful components are readily available at very low cost. Here we implement and evaluate the performance of the MDR algorithm on GPUs. Of primary interest are the time required for an epistasis analysis and the price to performance ratio of available solutions. We found that using MDR on GPUs consistently increased performance per machine over both a feature rich Java software package and a C++ cluster implementation. The performance of a GPU workstation running a GPU implementation reduces computation time by a factor of 160 compared to an 8-core workstation running the Java implementation on CPUs. This GPU workstation performs similarly to 150 cores running an optimized C++ implementation on a Beowulf cluster. Furthermore this GPU system provides extremely cost effective performance while leaving the CPU available for other tasks. The GPU workstation containing three GPUs costs $2000 while obtaining similar performance on a Beowulf cluster requires 150 CPU cores which, including the added infrastructure and support cost of the cluster system, cost approximately $82,500. Graphics hardware based computing provides a cost effective means to perform genetic analysis of epistasis using MDR on large datasets without the infrastructure of a computing cluster.

TU-AB-BRC-10: Modeling of Radiotherapy Linac Source Terms Using ARCHER Monte Carlo Code: Performance Comparison of GPU and MIC Computing Accelerators

DOE Office of Scientific and Technical Information (OSTI.GOV)

Liu, T; Lin, H; Xu, X

Purpose: (1) To perform phase space (PS) based source modeling for Tomotherapy and Varian TrueBeam 6 MV Linacs, (2) to examine the accuracy and performance of the ARCHER Monte Carlo code on a heterogeneous computing platform with Many Integrated Core coprocessors (MIC, aka Xeon Phi) and GPUs, and (3) to explore the software micro-optimization methods. Methods: The patient-specific source of Tomotherapy and Varian TrueBeam Linacs was modeled using the PS approach. For the helical Tomotherapy case, the PS data were calculated in our previous study (Su et al. 2014 41(7) Medical Physics). For the single-view Varian TrueBeam case, we analyticallymore » derived them from the raw patient-independent PS data in IAEA’s database, partial geometry information of the jaw and MLC as well as the fluence map. The phantom was generated from DICOM images. The Monte Carlo simulation was performed by ARCHER-MIC and GPU codes, which were benchmarked against a modified parallel DPM code. Software micro-optimization was systematically conducted, and was focused on SIMD vectorization of tight for-loops and data prefetch, with the ultimate goal of increasing 512-bit register utilization and reducing memory access latency. Results: Dose calculation was performed for two clinical cases, a Tomotherapy-based prostate cancer treatment and a TrueBeam-based left breast treatment. ARCHER was verified against the DPM code. The statistical uncertainty of the dose to the PTV was less than 1%. Using double-precision, the total wall time of the multithreaded CPU code on a X5650 CPU was 339 seconds for the Tomotherapy case and 131 seconds for the TrueBeam, while on 3 5110P MICs it was reduced to 79 and 59 seconds, respectively. The single-precision GPU code on a K40 GPU took 45 seconds for the Tomotherapy dose calculation. Conclusion: We have extended ARCHER, the MIC and GPU-based Monte Carlo dose engine to Tomotherapy and Truebeam dose calculations.« less

GPU-completeness: theory and implications

NASA Astrophysics Data System (ADS)

Lin, I.-Jong

2011-01-01

This paper formalizes a major insight into a class of algorithms that relate parallelism and performance. The purpose of this paper is to define a class of algorithms that trades off parallelism for quality of result (e.g. visual quality, compression rate), and we propose a similar method for algorithmic classification based on NP-Completeness techniques, applied toward parallel acceleration. We will define this class of algorithm as "GPU-Complete" and will postulate the necessary properties of the algorithms for admission into this class. We will also formally relate his algorithmic space and imaging algorithms space. This concept is based upon our experience in the print production area where GPUs (Graphic Processing Units) have shown a substantial cost/performance advantage within the context of HPdelivered enterprise services and commercial printing infrastructure. While CPUs and GPUs are converging in their underlying hardware and functional blocks, their system behaviors are clearly distinct in many ways: memory system design, programming paradigms, and massively parallel SIMD architecture. There are applications that are clearly suited to each architecture: for CPU: language compilation, word processing, operating systems, and other applications that are highly sequential in nature; for GPU: video rendering, particle simulation, pixel color conversion, and other problems clearly amenable to massive parallelization. While GPUs establishing themselves as a second, distinct computing architecture from CPUs, their end-to-end system cost/performance advantage in certain parts of computation inform the structure of algorithms and their efficient parallel implementations. While GPUs are merely one type of architecture for parallelization, we show that their introduction into the design space of printing systems demonstrate the trade-offs against competing multi-core, FPGA, and ASIC architectures. While each architecture has its own optimal application, we believe that the selection of architecture can be defined in terms of properties of GPU-Completeness. For a welldefined subset of algorithms, GPU-Completeness is intended to connect the parallelism, algorithms and efficient architectures into a unified framework to show that multiple layers of parallel implementation are guided by the same underlying trade-off.

An OpenACC-Based Unified Programming Model for Multi-accelerator Systems

DOE Office of Scientific and Technical Information (OSTI.GOV)

Kim, Jungwon; Lee, Seyong; Vetter, Jeffrey S

2015-01-01

This paper proposes a novel SPMD programming model of OpenACC. Our model integrates the different granularities of parallelism from vector-level parallelism to node-level parallelism into a single, unified model based on OpenACC. It allows programmers to write programs for multiple accelerators using a uniform programming model whether they are in shared or distributed memory systems. We implement a prototype of our model and evaluate its performance with a GPU-based supercomputer using three benchmark applications.

Octree-based, GPU implementation of a continuous cellular automaton for the simulation of complex, evolving surfaces

NASA Astrophysics Data System (ADS)

Ferrando, N.; Gosálvez, M. A.; Cerdá, J.; Gadea, R.; Sato, K.

2011-03-01

Presently, dynamic surface-based models are required to contain increasingly larger numbers of points and to propagate them over longer time periods. For large numbers of surface points, the octree data structure can be used as a balance between low memory occupation and relatively rapid access to the stored data. For evolution rules that depend on neighborhood states, extended simulation periods can be obtained by using simplified atomistic propagation models, such as the Cellular Automata (CA). This method, however, has an intrinsic parallel updating nature and the corresponding simulations are highly inefficient when performed on classical Central Processing Units (CPUs), which are designed for the sequential execution of tasks. In this paper, a series of guidelines is presented for the efficient adaptation of octree-based, CA simulations of complex, evolving surfaces into massively parallel computing hardware. A Graphics Processing Unit (GPU) is used as a cost-efficient example of the parallel architectures. For the actual simulations, we consider the surface propagation during anisotropic wet chemical etching of silicon as a computationally challenging process with a wide-spread use in microengineering applications. A continuous CA model that is intrinsically parallel in nature is used for the time evolution. Our study strongly indicates that parallel computations of dynamically evolving surfaces simulated using CA methods are significantly benefited by the incorporation of octrees as support data structures, substantially decreasing the overall computational time and memory usage.

An SDR-Based Real-Time Testbed for GNSS Adaptive Array Anti-Jamming Algorithms Accelerated by GPU

PubMed Central

Xu, Hailong; Cui, Xiaowei; Lu, Mingquan

2016-01-01

Nowadays, software-defined radio (SDR) has become a common approach to evaluate new algorithms. However, in the field of Global Navigation Satellite System (GNSS) adaptive array anti-jamming, previous work has been limited due to the high computational power demanded by adaptive algorithms, and often lack flexibility and configurability. In this paper, the design and implementation of an SDR-based real-time testbed for GNSS adaptive array anti-jamming accelerated by a Graphics Processing Unit (GPU) are documented. This testbed highlights itself as a feature-rich and extendible platform with great flexibility and configurability, as well as high computational performance. Both Space-Time Adaptive Processing (STAP) and Space-Frequency Adaptive Processing (SFAP) are implemented with a wide range of parameters. Raw data from as many as eight antenna elements can be processed in real-time in either an adaptive nulling or beamforming mode. To fully take advantage of the parallelism resource provided by the GPU, a batched method in programming is proposed. Tests and experiments are conducted to evaluate both the computational and anti-jamming performance. This platform can be used for research and prototyping, as well as a real product in certain applications. PMID:26978363

An SDR-Based Real-Time Testbed for GNSS Adaptive Array Anti-Jamming Algorithms Accelerated by GPU.

PubMed

Xu, Hailong; Cui, Xiaowei; Lu, Mingquan

2016-03-11

Nowadays, software-defined radio (SDR) has become a common approach to evaluate new algorithms. However, in the field of Global Navigation Satellite System (GNSS) adaptive array anti-jamming, previous work has been limited due to the high computational power demanded by adaptive algorithms, and often lack flexibility and configurability. In this paper, the design and implementation of an SDR-based real-time testbed for GNSS adaptive array anti-jamming accelerated by a Graphics Processing Unit (GPU) are documented. This testbed highlights itself as a feature-rich and extendible platform with great flexibility and configurability, as well as high computational performance. Both Space-Time Adaptive Processing (STAP) and Space-Frequency Adaptive Processing (SFAP) are implemented with a wide range of parameters. Raw data from as many as eight antenna elements can be processed in real-time in either an adaptive nulling or beamforming mode. To fully take advantage of the parallelism resource provided by the GPU, a batched method in programming is proposed. Tests and experiments are conducted to evaluate both the computational and anti-jamming performance. This platform can be used for research and prototyping, as well as a real product in certain applications.

The Pore-scale modeling of multiphase flows in reservoir rocks using the lattice Boltzmann method

NASA Astrophysics Data System (ADS)

Mu, Y.; Baldwin, C. H.; Toelke, J.; Grader, A.

2011-12-01

Digital rock physics (DRP) is a new technology to compute the physical and fluid flow properties of reservoir rocks. In this approach, pore scale images of the porous rock are obtained and processed to create highly accurate 3D digital rock sample, and then the rock properties are evaluated by advanced numerical methods at the pore scale. Ingrain's DRP technology is a breakthrough for oil and gas companies that need large volumes of accurate results faster than the current special core analysis (SCAL) laboratories can normally deliver. In this work, we compute the multiphase fluid flow properties of 3D digital rocks using D3Q19 immiscible LBM with two relaxation times (TRT). For efficient implementation on GPU, we improved and reformulated color-gradient model proposed by Gunstensen and Rothmann. Furthermore, we only use one-lattice with the sparse data structure: only allocate memory for pore nodes on GPU. We achieved more than 100 million fluid lattice updates per second (MFLUPS) for two-phase LBM on single Fermi-GPU and high parallel efficiency on Multi-GPUs. We present and discuss our simulation results of important two-phase fluid flow properties, such as capillary pressure and relative permeabilities. We also investigate the effects of resolution and wettability on multiphase flows. Comparison of direct measurement results with the LBM-based simulations shows practical ability of DRP to predict two-phase flow properties of reservoir rock.

GPUbased, Microsecond Latency, HectoChannel MIMO Feedback Control of Magnetically Confined Plasmas

NASA Astrophysics Data System (ADS)

Rath, Nikolaus

Feedback control has become a crucial tool in the research on magnetic confinement of plasmas for achieving controlled nuclear fusion. This thesis presents a novel plasma feedback control system that, for the first time, employs a Graphics Processing Unit (GPU) for microsecond-latency, real-time control computations. This novel application area for GPU computing is opened up by a new system architecture that is optimized for low-latency computations on less than kilobyte sized data samples as they occur in typical plasma control algorithms. In contrast to traditional GPU computing approaches that target complex, high-throughput computations with massive amounts of data, the architecture presented in this thesis uses the GPU as the primary processing unit rather than as an auxiliary of the CPU, and data is transferred from A-D/D-A converters directly into GPU memory using peer-to-peer PCI Express transfers. The described design has been implemented in a new, GPU-based control system for the High-Beta Tokamak - Extended Pulse (HBT-EP) device. The system is built from commodity hardware and uses an NVIDIA GeForce GPU and D-TACQ A-D/D-A converters providing a total of 96 input and 64 output channels. The system is able to run with sampling periods down to 4 μs and latencies down to 8 μs. The GPU provides a total processing power of 1.5 x 1012 floating point operations per second. To illustrate the performance and versatility of both the general architecture and concrete implementation, a new control algorithm has been developed. The algorithm is designed for the control of multiple rotating magnetic perturbations in situations where the plasma equilibrium is not known exactly and features an adaptive system model: instead of requiring the rotation frequencies and growth rates embedded in the system model to be set a priori, the adaptive algorithm derives these parameters from the evolution of the perturbation amplitudes themselves. This results in non-linear control computations with high computational demands, but is handled easily by the GPU based system. Both digital processing latency and an arbitrary multi-pole response of amplifiers and control coils is fully taken into account for the generation of control signals. To separate sensor signals into perturbed and equilibrium components without knowledge of the equilibrium fields, a new separation method based on biorthogonal decomposition is introduced and used to derive a filter that performs the separation in real-time. The control algorithm has been implemented and tested on the new, GPU-based feedback control system of the HBT-EP tokamak. In this instance, the algorithm was set up to control four rotating n = 1 perturbations at different poloidal angles. The perturbations were treated as coupled in frequency but independent in amplitude and phase, so that the system effectively controls a helical n = 1 perturbation with unknown poloidal spectrum. Depending on the plasma's edge safety factor and rotation frequency, the control system is shown to be able to suppress the amplitude of the dominant 8 kHz mode by up to 60% or amplify the saturated amplitude by a factor of up to two. Intermediate feedback phases combine suppression and amplification with a speed up or slow down of the mode rotation frequency. Increasing feedback gain results in the excitation of an additional, slowly rotating 1.4 kHz mode without further effects on the 8 kHz mode. The feedback performance is found to exceed previous results obtained with an FPGA- and Kalman-filter based control system without requiring any tuning of system model parameters. Experimental results are compared with simulations based on a combination of the Boozer surface current model and the Fitzpatrick-Aydemir model. Within the subset of phenomena that can be represented by the model as well as determined experimentally, qualitative agreement is found.

Data Parallel Bin-Based Indexing for Answering Queries on Multi-Core Architectures

DOE Office of Scientific and Technical Information (OSTI.GOV)

Gosink, Luke; Wu, Kesheng; Bethel, E. Wes

2009-06-02

The multi-core trend in CPUs and general purpose graphics processing units (GPUs) offers new opportunities for the database community. The increase of cores at exponential rates is likely to affect virtually every server and client in the coming decade, and presents database management systems with a huge, compelling disruption that will radically change how processing is done. This paper presents a new parallel indexing data structure for answering queries that takes full advantage of the increasing thread-level parallelism emerging in multi-core architectures. In our approach, our Data Parallel Bin-based Index Strategy (DP-BIS) first bins the base data, and then partitionsmore » and stores the values in each bin as a separate, bin-based data cluster. In answering a query, the procedures for examining the bin numbers and the bin-based data clusters offer the maximum possible level of concurrency; each record is evaluated by a single thread and all threads are processed simultaneously in parallel. We implement and demonstrate the effectiveness of DP-BIS on two multi-core architectures: a multi-core CPU and a GPU. The concurrency afforded by DP-BIS allows us to fully utilize the thread-level parallelism provided by each architecture--for example, our GPU-based DP-BIS implementation simultaneously evaluates over 12,000 records with an equivalent number of concurrently executing threads. In comparing DP-BIS's performance across these architectures, we show that the GPU-based DP-BIS implementation requires significantly less computation time to answer a query than the CPU-based implementation. We also demonstrate in our analysis that DP-BIS provides better overall performance than the commonly utilized CPU and GPU-based projection index. Finally, due to data encoding, we show that DP-BIS accesses significantly smaller amounts of data than index strategies that operate solely on a column's base data; this smaller data footprint is critical for parallel processors that possess limited memory resources (e.g., GPUs).« less

A GPU-accelerated and Monte Carlo-based intensity modulated proton therapy optimization system.

PubMed

Ma, Jiasen; Beltran, Chris; Seum Wan Chan Tseung, Hok; Herman, Michael G

2014-12-01

Conventional spot scanning intensity modulated proton therapy (IMPT) treatment planning systems (TPSs) optimize proton spot weights based on analytical dose calculations. These analytical dose calculations have been shown to have severe limitations in heterogeneous materials. Monte Carlo (MC) methods do not have these limitations; however, MC-based systems have been of limited clinical use due to the large number of beam spots in IMPT and the extremely long calculation time of traditional MC techniques. In this work, the authors present a clinically applicable IMPT TPS that utilizes a very fast MC calculation. An in-house graphics processing unit (GPU)-based MC dose calculation engine was employed to generate the dose influence map for each proton spot. With the MC generated influence map, a modified least-squares optimization method was used to achieve the desired dose volume histograms (DVHs). The intrinsic CT image resolution was adopted for voxelization in simulation and optimization to preserve spatial resolution. The optimizations were computed on a multi-GPU framework to mitigate the memory limitation issues for the large dose influence maps that resulted from maintaining the intrinsic CT resolution. The effects of tail cutoff and starting condition were studied and minimized in this work. For relatively large and complex three-field head and neck cases, i.e., >100,000 spots with a target volume of ∼ 1000 cm(3) and multiple surrounding critical structures, the optimization together with the initial MC dose influence map calculation was done in a clinically viable time frame (less than 30 min) on a GPU cluster consisting of 24 Nvidia GeForce GTX Titan cards. The in-house MC TPS plans were comparable to a commercial TPS plans based on DVH comparisons. A MC-based treatment planning system was developed. The treatment planning can be performed in a clinically viable time frame on a hardware system costing around 45,000 dollars. The fast calculation and optimization make the system easily expandable to robust and multicriteria optimization.

hybrid\\scriptsize{{MANTIS}}: a CPU-GPU Monte Carlo method for modeling indirect x-ray detectors with columnar scintillators

NASA Astrophysics Data System (ADS)

Sharma, Diksha; Badal, Andreu; Badano, Aldo

2012-04-01

The computational modeling of medical imaging systems often requires obtaining a large number of simulated images with low statistical uncertainty which translates into prohibitive computing times. We describe a novel hybrid approach for Monte Carlo simulations that maximizes utilization of CPUs and GPUs in modern workstations. We apply the method to the modeling of indirect x-ray detectors using a new and improved version of the code \\scriptsize{{MANTIS}}, an open source software tool used for the Monte Carlo simulations of indirect x-ray imagers. We first describe a GPU implementation of the physics and geometry models in fast\\scriptsize{{DETECT}}2 (the optical transport model) and a serial CPU version of the same code. We discuss its new features like on-the-fly column geometry and columnar crosstalk in relation to the \\scriptsize{{MANTIS}} code, and point out areas where our model provides more flexibility for the modeling of realistic columnar structures in large area detectors. Second, we modify \\scriptsize{{PENELOPE}} (the open source software package that handles the x-ray and electron transport in \\scriptsize{{MANTIS}}) to allow direct output of location and energy deposited during x-ray and electron interactions occurring within the scintillator. This information is then handled by optical transport routines in fast\\scriptsize{{DETECT}}2. A load balancer dynamically allocates optical transport showers to the GPU and CPU computing cores. Our hybrid\\scriptsize{{MANTIS}} approach achieves a significant speed-up factor of 627 when compared to \\scriptsize{{MANTIS}} and of 35 when compared to the same code running only in a CPU instead of a GPU. Using hybrid\\scriptsize{{MANTIS}}, we successfully hide hours of optical transport time by running it in parallel with the x-ray and electron transport, thus shifting the computational bottleneck from optical to x-ray transport. The new code requires much less memory than \\scriptsize{{MANTIS}} and, as a result, allows us to efficiently simulate large area detectors.

A performance model for GPUs with caches

DOE PAGES

Dao, Thanh Tuan; Kim, Jungwon; Seo, Sangmin; ...

2014-06-24

To exploit the abundant computational power of the world's fastest supercomputers, an even workload distribution to the typically heterogeneous compute devices is necessary. While relatively accurate performance models exist for conventional CPUs, accurate performance estimation models for modern GPUs do not exist. This paper presents two accurate models for modern GPUs: a sampling-based linear model, and a model based on machine-learning (ML) techniques which improves the accuracy of the linear model and is applicable to modern GPUs with and without caches. We first construct the sampling-based linear model to predict the runtime of an arbitrary OpenCL kernel. Based on anmore » analysis of NVIDIA GPUs' scheduling policies we determine the earliest sampling points that allow an accurate estimation. The linear model cannot capture well the significant effects that memory coalescing or caching as implemented in modern GPUs have on performance. We therefore propose a model based on ML techniques that takes several compiler-generated statistics about the kernel as well as the GPU's hardware performance counters as additional inputs to obtain a more accurate runtime performance estimation for modern GPUs. We demonstrate the effectiveness and broad applicability of the model by applying it to three different NVIDIA GPU architectures and one AMD GPU architecture. On an extensive set of OpenCL benchmarks, on average, the proposed model estimates the runtime performance with less than 7 percent error for a second-generation GTX 280 with no on-chip caches and less than 5 percent for the Fermi-based GTX 580 with hardware caches. On the Kepler-based GTX 680, the linear model has an error of less than 10 percent. On an AMD GPU architecture, Radeon HD 6970, the model estimates with 8 percent of error rates. As a result, the proposed technique outperforms existing models by a factor of 5 to 6 in terms of accuracy.« less

Accelerated Prediction of the Polar Ice and Global Ocean (APPIGO)

DTIC Science & Technology

2015-09-30

and so HYCOM was 10x slower on the K20X Keplers than on the 16-Core AMDs alone. The initial lack of performance was not a surprise. Our goal was to...MPI ranks to take advantage of the Hyper-Q capabilities on newer Kepler architectures. Hyper-Q allows multiple GPU kernels originating from

DOE Office of Scientific and Technical Information (OSTI.GOV)

Perumalla, Kalyan S.; Yoginath, Srikanth B.

Problems such as fault tolerance and scalable synchronization can be efficiently solved using reversibility of applications. Making applications reversible by relying on computation rather than on memory is ideal for large scale parallel computing, especially for the next generation of supercomputers in which memory is expensive in terms of latency, energy, and price. In this direction, a case study is presented here in reversing a computational core, namely, Basic Linear Algebra Subprograms, which is widely used in scientific applications. A new Reversible BLAS (RBLAS) library interface has been designed, and a prototype has been implemented with two modes: (1) amore » memory-mode in which reversibility is obtained by checkpointing to memory in forward and restoring from memory in reverse, and (2) a computational-mode in which nothing is saved in the forward, but restoration is done entirely via inverse computation in reverse. The article is focused on detailed performance benchmarking to evaluate the runtime dynamics and performance effects, comparing reversible computation with checkpointing on both traditional CPU platforms and recent GPU accelerator platforms. For BLAS Level-1 subprograms, data indicates over an order of magnitude better speed of reversible computation compared to checkpointing. For BLAS Level-2 and Level-3, a more complex tradeoff is observed between reversible computation and checkpointing, depending on computational and memory complexities of the subprograms.« less

High-Speed GPU-Based Fully Three-Dimensional Diffuse Optical Tomographic System

PubMed Central

Saikia, Manob Jyoti; Kanhirodan, Rajan; Mohan Vasu, Ram

2014-01-01

We have developed a graphics processor unit (GPU-) based high-speed fully 3D system for diffuse optical tomography (DOT). The reduction in execution time of 3D DOT algorithm, a severely ill-posed problem, is made possible through the use of (1) an algorithmic improvement that uses Broyden approach for updating the Jacobian matrix and thereby updating the parameter matrix and (2) the multinode multithreaded GPU and CUDA (Compute Unified Device Architecture) software architecture. Two different GPU implementations of DOT programs are developed in this study: (1) conventional C language program augmented by GPU CUDA and CULA routines (C GPU), (2) MATLAB program supported by MATLAB parallel computing toolkit for GPU (MATLAB GPU). The computation time of the algorithm on host CPU and the GPU system is presented for C and Matlab implementations. The forward computation uses finite element method (FEM) and the problem domain is discretized into 14610, 30823, and 66514 tetrahedral elements. The reconstruction time, so achieved for one iteration of the DOT reconstruction for 14610 elements, is 0.52 seconds for a C based GPU program for 2-plane measurements. The corresponding MATLAB based GPU program took 0.86 seconds. The maximum number of reconstructed frames so achieved is 2 frames per second. PMID:24891848

High-Speed GPU-Based Fully Three-Dimensional Diffuse Optical Tomographic System.

PubMed

Saikia, Manob Jyoti; Kanhirodan, Rajan; Mohan Vasu, Ram

2014-01-01

We have developed a graphics processor unit (GPU-) based high-speed fully 3D system for diffuse optical tomography (DOT). The reduction in execution time of 3D DOT algorithm, a severely ill-posed problem, is made possible through the use of (1) an algorithmic improvement that uses Broyden approach for updating the Jacobian matrix and thereby updating the parameter matrix and (2) the multinode multithreaded GPU and CUDA (Compute Unified Device Architecture) software architecture. Two different GPU implementations of DOT programs are developed in this study: (1) conventional C language program augmented by GPU CUDA and CULA routines (C GPU), (2) MATLAB program supported by MATLAB parallel computing toolkit for GPU (MATLAB GPU). The computation time of the algorithm on host CPU and the GPU system is presented for C and Matlab implementations. The forward computation uses finite element method (FEM) and the problem domain is discretized into 14610, 30823, and 66514 tetrahedral elements. The reconstruction time, so achieved for one iteration of the DOT reconstruction for 14610 elements, is 0.52 seconds for a C based GPU program for 2-plane measurements. The corresponding MATLAB based GPU program took 0.86 seconds. The maximum number of reconstructed frames so achieved is 2 frames per second.

MIGS-GPU: Microarray Image Gridding and Segmentation on the GPU.

PubMed

Katsigiannis, Stamos; Zacharia, Eleni; Maroulis, Dimitris

2017-05-01

Complementary DNA (cDNA) microarray is a powerful tool for simultaneously studying the expression level of thousands of genes. Nevertheless, the analysis of microarray images remains an arduous and challenging task due to the poor quality of the images that often suffer from noise, artifacts, and uneven background. In this study, the MIGS-GPU [Microarray Image Gridding and Segmentation on Graphics Processing Unit (GPU)] software for gridding and segmenting microarray images is presented. MIGS-GPU's computations are performed on the GPU by means of the compute unified device architecture (CUDA) in order to achieve fast performance and increase the utilization of available system resources. Evaluation on both real and synthetic cDNA microarray images showed that MIGS-GPU provides better performance than state-of-the-art alternatives, while the proposed GPU implementation achieves significantly lower computational times compared to the respective CPU approaches. Consequently, MIGS-GPU can be an advantageous and useful tool for biomedical laboratories, offering a user-friendly interface that requires minimum input in order to run.

Extending the length and time scales of Gram–Schmidt Lyapunov vector computations

DOE Office of Scientific and Technical Information (OSTI.GOV)

Costa, Anthony B., E-mail: acosta@northwestern.edu; Green, Jason R., E-mail: jason.green@umb.edu; Department of Chemistry, University of Massachusetts Boston, Boston, MA 02125

Lyapunov vectors have found growing interest recently due to their ability to characterize systems out of thermodynamic equilibrium. The computation of orthogonal Gram–Schmidt vectors requires multiplication and QR decomposition of large matrices, which grow as N{sup 2} (with the particle count). This expense has limited such calculations to relatively small systems and short time scales. Here, we detail two implementations of an algorithm for computing Gram–Schmidt vectors. The first is a distributed-memory message-passing method using Scalapack. The second uses the newly-released MAGMA library for GPUs. We compare the performance of both codes for Lennard–Jones fluids from N=100 to 1300 betweenmore » Intel Nahalem/Infiniband DDR and NVIDIA C2050 architectures. To our best knowledge, these are the largest systems for which the Gram–Schmidt Lyapunov vectors have been computed, and the first time their calculation has been GPU-accelerated. We conclude that Lyapunov vector calculations can be significantly extended in length and time by leveraging the power of GPU-accelerated linear algebra.« less

Heterogeneous compute in computer vision: OpenCL in OpenCV

NASA Astrophysics Data System (ADS)

Gasparakis, Harris

2014-02-01

We explore the relevance of Heterogeneous System Architecture (HSA) in Computer Vision, both as a long term vision, and as a near term emerging reality via the recently ratified OpenCL 2.0 Khronos standard. After a brief review of OpenCL 1.2 and 2.0, including HSA features such as Shared Virtual Memory (SVM) and platform atomics, we identify what genres of Computer Vision workloads stand to benefit by leveraging those features, and we suggest a new mental framework that replaces GPU compute with hybrid HSA APU compute. As a case in point, we discuss, in some detail, popular object recognition algorithms (part-based models), emphasizing the interplay and concurrent collaboration between the GPU and CPU. We conclude by describing how OpenCL has been incorporated in OpenCV, a popular open source computer vision library, emphasizing recent work on the Transparent API, to appear in OpenCV 3.0, which unifies the native CPU and OpenCL execution paths under a single API, allowing the same code to execute either on CPU or on a OpenCL enabled device, without even recompiling.

Object tracking mask-based NLUT on GPUs for real-time generation of holographic videos of three-dimensional scenes.

PubMed

Kwon, M-W; Kim, S-C; Yoon, S-E; Ho, Y-S; Kim, E-S

2015-02-09

A new object tracking mask-based novel-look-up-table (OTM-NLUT) method is proposed and implemented on graphics-processing-units (GPUs) for real-time generation of holographic videos of three-dimensional (3-D) scenes. Since the proposed method is designed to be matched with software and memory structures of the GPU, the number of compute-unified-device-architecture (CUDA) kernel function calls and the computer-generated hologram (CGH) buffer size of the proposed method have been significantly reduced. It therefore results in a great increase of the computational speed of the proposed method and enables real-time generation of CGH patterns of 3-D scenes. Experimental results show that the proposed method can generate 31.1 frames of Fresnel CGH patterns with 1,920 × 1,080 pixels per second, on average, for three test 3-D video scenarios with 12,666 object points on three GPU boards of NVIDIA GTX TITAN, and confirm the feasibility of the proposed method in the practical application of electro-holographic 3-D displays.

Building a Terabyte Memory Bandwidth Compute Node with Four Consumer Electronics GPUs

NASA Astrophysics Data System (ADS)

Omlin, Samuel; Räss, Ludovic; Podladchikov, Yuri

2014-05-01

GPUs released for consumer electronics are generally built with the same chip architectures as the GPUs released for professional usage. With regards to scientific computing, there are no obvious important differences in functionality or performance between the two types of releases, yet the price can differ up to one order of magnitude. For example, the consumer electronics release of the most recent NVIDIA Kepler architecture (GK110), named GeForce GTX TITAN, performed equally well in conducted memory bandwidth tests as the professional release, named Tesla K20; the consumer electronics release costs about one third of the professional release. We explain how to design and assemble a well adjusted computer with four high-end consumer electronics GPUs (GeForce GTX TITAN) combining more than 1 terabyte/s memory bandwidth. We compare the system's performance and precision with the one of hardware released for professional usage. The system can be used as a powerful workstation for scientific computing or as a compute node in a home-built GPU cluster.

Quasi-real-time end-to-end simulations of ELT-scale adaptive optics systems on GPUs

NASA Astrophysics Data System (ADS)

Gratadour, Damien

2011-09-01

Our team has started the development of a code dedicated to GPUs for the simulation of AO systems at the E-ELT scale. It uses the CUDA toolkit and an original binding to Yorick (an open source interpreted language) to provide the user with a comprehensive interface. In this paper we present the first performance analysis of our simulation code, showing its ability to provide Shack-Hartmann (SH) images and measurements at the kHz scale for VLT-sized AO system and in quasi-real-time (up to 70 Hz) for ELT-sized systems on a single top-end GPU. The simulation code includes multiple layers atmospheric turbulence generation, ray tracing through these layers, image formation at the focal plane of every sub-apertures of a SH sensor using either natural or laser guide stars and centroiding on these images using various algorithms. Turbulence is generated on-the-fly giving the ability to simulate hours of observations without the need of loading extremely large phase screens in the global memory. Because of its performance this code additionally provides the unique ability to test real-time controllers for future AO systems under nominal conditions.

Parallel algorithm for solving Kepler’s equation on Graphics Processing Units: Application to analysis of Doppler exoplanet searches

NASA Astrophysics Data System (ADS)

Ford, Eric B.

2009-05-01

We present the results of a highly parallel Kepler equation solver using the Graphics Processing Unit (GPU) on a commercial nVidia GeForce 280GTX and the "Compute Unified Device Architecture" (CUDA) programming environment. We apply this to evaluate a goodness-of-fit statistic (e.g., χ2) for Doppler observations of stars potentially harboring multiple planetary companions (assuming negligible planet-planet interactions). Given the high-dimensionality of the model parameter space (at least five dimensions per planet), a global search is extremely computationally demanding. We expect that the underlying Kepler solver and model evaluator will be combined with a wide variety of more sophisticated algorithms to provide efficient global search, parameter estimation, model comparison, and adaptive experimental design for radial velocity and/or astrometric planet searches. We tested multiple implementations using single precision, double precision, pairs of single precision, and mixed precision arithmetic. We find that the vast majority of computations can be performed using single precision arithmetic, with selective use of compensated summation for increased precision. However, standard single precision is not adequate for calculating the mean anomaly from the time of observation and orbital period when evaluating the goodness-of-fit for real planetary systems and observational data sets. Using all double precision, our GPU code outperforms a similar code using a modern CPU by a factor of over 60. Using mixed precision, our GPU code provides a speed-up factor of over 600, when evaluating nsys > 1024 models planetary systems each containing npl = 4 planets and assuming nobs = 256 observations of each system. We conclude that modern GPUs also offer a powerful tool for repeatedly evaluating Kepler's equation and a goodness-of-fit statistic for orbital models when presented with a large parameter space.

SU-D-BRD-03: A Gateway for GPU Computing in Cancer Radiotherapy Research

DOE Office of Scientific and Technical Information (OSTI.GOV)

Jia, X; Folkerts, M; Shi, F

Purpose: Graphics Processing Unit (GPU) has become increasingly important in radiotherapy. However, it is still difficult for general clinical researchers to access GPU codes developed by other researchers, and for developers to objectively benchmark their codes. Moreover, it is quite often to see repeated efforts spent on developing low-quality GPU codes. The goal of this project is to establish an infrastructure for testing GPU codes, cross comparing them, and facilitating code distributions in radiotherapy community. Methods: We developed a system called Gateway for GPU Computing in Cancer Radiotherapy Research (GCR2). A number of GPU codes developed by our group andmore » other developers can be accessed via a web interface. To use the services, researchers first upload their test data or use the standard data provided by our system. Then they can select the GPU device on which the code will be executed. Our system offers all mainstream GPU hardware for code benchmarking purpose. After the code running is complete, the system automatically summarizes and displays the computing results. We also released a SDK to allow the developers to build their own algorithm implementation and submit their binary codes to the system. The submitted code is then systematically benchmarked using a variety of GPU hardware and representative data provided by our system. The developers can also compare their codes with others and generate benchmarking reports. Results: It is found that the developed system is fully functioning. Through a user-friendly web interface, researchers are able to test various GPU codes. Developers also benefit from this platform by comprehensively benchmarking their codes on various GPU platforms and representative clinical data sets. Conclusion: We have developed an open platform allowing the clinical researchers and developers to access the GPUs and GPU codes. This development will facilitate the utilization of GPU in radiation therapy field.« less

GPU: the biggest key processor for AI and parallel processing

NASA Astrophysics Data System (ADS)

Baji, Toru

2017-07-01

Two types of processors exist in the market. One is the conventional CPU and the other is Graphic Processor Unit (GPU). Typical CPU is composed of 1 to 8 cores while GPU has thousands of cores. CPU is good for sequential processing, while GPU is good to accelerate software with heavy parallel executions. GPU was initially dedicated for 3D graphics. However from 2006, when GPU started to apply general-purpose cores, it was noticed that this architecture can be used as a general purpose massive-parallel processor. NVIDIA developed a software framework Compute Unified Device Architecture (CUDA) that make it possible to easily program the GPU for these application. With CUDA, GPU started to be used in workstations and supercomputers widely. Recently two key technologies are highlighted in the industry. The Artificial Intelligence (AI) and Autonomous Driving Cars. AI requires a massive parallel operation to train many-layers of neural networks. With CPU alone, it was impossible to finish the training in a practical time. The latest multi-GPU system with P100 makes it possible to finish the training in a few hours. For the autonomous driving cars, TOPS class of performance is required to implement perception, localization, path planning processing and again SoC with integrated GPU will play a key role there. In this paper, the evolution of the GPU which is one of the biggest commercial devices requiring state-of-the-art fabrication technology will be introduced. Also overview of the GPU demanding key application like the ones described above will be introduced.

Depth compensating calculation method of computer-generated holograms using symmetry and similarity of zone plates

NASA Astrophysics Data System (ADS)

Wei, Hui; Gong, Guanghong; Li, Ni

2017-10-01

Computer-generated hologram (CGH) is a promising 3D display technology while it is challenged by heavy computation load and vast memory requirement. To solve these problems, a depth compensating CGH calculation method based on symmetry and similarity of zone plates is proposed and implemented on graphics processing unit (GPU). An improved LUT method is put forward to compute the distances between object points and hologram pixels in the XY direction. The concept of depth compensating factor is defined and used for calculating the holograms of points with different depth positions instead of layer-based methods. The proposed method is suitable for arbitrary sampling objects with lower memory usage and higher computational efficiency compared to other CGH methods. The effectiveness of the proposed method is validated by numerical and optical experiments.

Development of a GPU Compatible Version of the Fast Radiation Code RRTMG

NASA Astrophysics Data System (ADS)

Iacono, M. J.; Mlawer, E. J.; Berthiaume, D.; Cady-Pereira, K. E.; Suarez, M.; Oreopoulos, L.; Lee, D.

2012-12-01

The absorption of solar radiation and emission/absorption of thermal radiation are crucial components of the physics that drive Earth's climate and weather. Therefore, accurate radiative transfer calculations are necessary for realistic climate and weather simulations. Efficient radiation codes have been developed for this purpose, but their accuracy requirements still necessitate that as much as 30% of the computational time of a GCM is spent computing radiative fluxes and heating rates. The overall computational expense constitutes a limitation on a GCM's predictive ability if it becomes an impediment to adding new physics to or increasing the spatial and/or vertical resolution of the model. The emergence of Graphics Processing Unit (GPU) technology, which will allow the parallel computation of multiple independent radiative calculations in a GCM, will lead to a fundamental change in the competition between accuracy and speed. Processing time previously consumed by radiative transfer will now be available for the modeling of other processes, such as physics parameterizations, without any sacrifice in the accuracy of the radiative transfer. Furthermore, fast radiation calculations can be performed much more frequently and will allow the modeling of radiative effects of rapid changes in the atmosphere. The fast radiation code RRTMG, developed at Atmospheric and Environmental Research (AER), is utilized operationally in many dynamical models throughout the world. We will present the results from the first stage of an effort to create a version of the RRTMG radiation code designed to run efficiently in a GPU environment. This effort will focus on the RRTMG implementation in GEOS-5. RRTMG has an internal pseudo-spectral vector of length of order 100 that, when combined with the much greater length of the global horizontal grid vector from which the radiation code is called in GEOS-5, makes RRTMG/GEOS-5 particularly suited to achieving a significant speed improvement through GPU technology. This large number of independent cases will allow us to take full advantage of the computational power of the latest GPUs, ensuring that all thread cores in the GPU remain active, a key criterion for obtaining significant speedup. The CUDA (Compute Unified Device Architecture) Fortran compiler developed by PGI and Nvidia will allow us to construct this parallel implementation on the GPU while remaining in the Fortran language. This implementation will scale very well across various CUDA-supported GPUs such as the recently released Fermi Nvidia cards. We will present the computational speed improvements of the GPU-compatible code relative to the standard CPU-based RRTMG with respect to a very large and diverse suite of atmospheric profiles. This suite will also be utilized to demonstrate the minimal impact of the code restructuring on the accuracy of radiation calculations. The GPU-compatible version of RRTMG will be directly applicable to future versions of GEOS-5, but it is also likely to provide significant associated benefits for other GCMs that employ RRTMG.

Software beamforming: comparison between a phased array and synthetic transmit aperture.

PubMed

Li, Yen-Feng; Li, Pai-Chi

2011-04-01

The data-transfer and computation requirements are compared between software-based beamforming using a phased array (PA) and a synthetic transmit aperture (STA). The advantages of a software-based architecture are reduced system complexity and lower hardware cost. Although this architecture can be implemented using commercial CPUs or GPUs, the high computation and data-transfer requirements limit its real-time beamforming performance. In particular, transferring the raw rf data from the front-end subsystem to the software back-end remains challenging with current state-of-the-art electronics technologies, which offset the cost advantage of the software back end. This study investigated the tradeoff between the data-transfer and computation requirements. Two beamforming methods based on a PA and STA, respectively, were used: the former requires a higher data transfer rate and the latter requires more memory operations. The beamformers were implemente;d in an NVIDIA GeForce GTX 260 GPU and an Intel core i7 920 CPU. The frame rate of PA beamforming was 42 fps with a 128-element array transducer, with 2048 samples per firing and 189 beams per image (with a 95 MB/frame data-transfer requirement). The frame rate of STA beamforming was 40 fps with 16 firings per image (with an 8 MB/frame data-transfer requirement). Both approaches achieved real-time beamforming performance but each had its own bottleneck. On the one hand, the required data-transfer speed was considerably reduced in STA beamforming, whereas this required more memory operations, which limited the overall computation time. The advantages of the GPU approach over the CPU approach were clearly demonstrated.

Clinical implementation of a GPU-based simplified Monte Carlo method for a treatment planning system of proton beam therapy.

PubMed

Kohno, R; Hotta, K; Nishioka, S; Matsubara, K; Tansho, R; Suzuki, T

2011-11-21

We implemented the simplified Monte Carlo (SMC) method on graphics processing unit (GPU) architecture under the computer-unified device architecture platform developed by NVIDIA. The GPU-based SMC was clinically applied for four patients with head and neck, lung, or prostate cancer. The results were compared to those obtained by a traditional CPU-based SMC with respect to the computation time and discrepancy. In the CPU- and GPU-based SMC calculations, the estimated mean statistical errors of the calculated doses in the planning target volume region were within 0.5% rms. The dose distributions calculated by the GPU- and CPU-based SMCs were similar, within statistical errors. The GPU-based SMC showed 12.30-16.00 times faster performance than the CPU-based SMC. The computation time per beam arrangement using the GPU-based SMC for the clinical cases ranged 9-67 s. The results demonstrate the successful application of the GPU-based SMC to a clinical proton treatment planning.

GPU Optimizations for a Production Molecular Docking Code*

PubMed Central

Landaverde, Raphael; Herbordt, Martin C.

2015-01-01

Modeling molecular docking is critical to both understanding life processes and designing new drugs. In previous work we created the first published GPU-accelerated docking code (PIPER) which achieved a roughly 5× speed-up over a contemporaneous 4 core CPU. Advances in GPU architecture and in the CPU code, however, have since reduced this relalative performance by a factor of 10. In this paper we describe the upgrade of GPU PIPER. This required an entire rewrite, including algorithm changes and moving most remaining non-accelerated CPU code onto the GPU. The result is a 7× improvement in GPU performance and a 3.3× speedup over the CPU-only code. We find that this difference in time is almost entirely due to the difference in run times of the 3D FFT library functions on CPU (MKL) and GPU (cuFFT), respectively. The GPU code has been integrated into the ClusPro docking server which has over 4000 active users. PMID:26594667

GPU Optimizations for a Production Molecular Docking Code.

PubMed

Landaverde, Raphael; Herbordt, Martin C

2014-09-01

Modeling molecular docking is critical to both understanding life processes and designing new drugs. In previous work we created the first published GPU-accelerated docking code (PIPER) which achieved a roughly 5× speed-up over a contemporaneous 4 core CPU. Advances in GPU architecture and in the CPU code, however, have since reduced this relalative performance by a factor of 10. In this paper we describe the upgrade of GPU PIPER. This required an entire rewrite, including algorithm changes and moving most remaining non-accelerated CPU code onto the GPU. The result is a 7× improvement in GPU performance and a 3.3× speedup over the CPU-only code. We find that this difference in time is almost entirely due to the difference in run times of the 3D FFT library functions on CPU (MKL) and GPU (cuFFT), respectively. The GPU code has been integrated into the ClusPro docking server which has over 4000 active users.

Multi-GPU hybrid programming accelerated three-dimensional phase-field model in binary alloy

NASA Astrophysics Data System (ADS)

Zhu, Changsheng; Liu, Jieqiong; Zhu, Mingfang; Feng, Li

2018-03-01

In the process of dendritic growth simulation, the computational efficiency and the problem scales have extremely important influence on simulation efficiency of three-dimensional phase-field model. Thus, seeking for high performance calculation method to improve the computational efficiency and to expand the problem scales has a great significance to the research of microstructure of the material. A high performance calculation method based on MPI+CUDA hybrid programming model is introduced. Multi-GPU is used to implement quantitative numerical simulations of three-dimensional phase-field model in binary alloy under the condition of multi-physical processes coupling. The acceleration effect of different GPU nodes on different calculation scales is explored. On the foundation of multi-GPU calculation model that has been introduced, two optimization schemes, Non-blocking communication optimization and overlap of MPI and GPU computing optimization, are proposed. The results of two optimization schemes and basic multi-GPU model are compared. The calculation results show that the use of multi-GPU calculation model can improve the computational efficiency of three-dimensional phase-field obviously, which is 13 times to single GPU, and the problem scales have been expanded to 8193. The feasibility of two optimization schemes is shown, and the overlap of MPI and GPU computing optimization has better performance, which is 1.7 times to basic multi-GPU model, when 21 GPUs are used.

A local time stepping algorithm for GPU-accelerated 2D shallow water models

NASA Astrophysics Data System (ADS)

Dazzi, Susanna; Vacondio, Renato; Dal Palù, Alessandro; Mignosa, Paolo

2018-01-01

In the simulation of flooding events, mesh refinement is often required to capture local bathymetric features and/or to detail areas of interest; however, if an explicit finite volume scheme is adopted, the presence of small cells in the domain can restrict the allowable time step due to the stability condition, thus reducing the computational efficiency. With the aim of overcoming this problem, the paper proposes the application of a Local Time Stepping (LTS) strategy to a GPU-accelerated 2D shallow water numerical model able to handle non-uniform structured meshes. The algorithm is specifically designed to exploit the computational capability of GPUs, minimizing the overheads associated with the LTS implementation. The results of theoretical and field-scale test cases show that the LTS model guarantees appreciable reductions in the execution time compared to the traditional Global Time Stepping strategy, without compromising the solution accuracy.

Binary mesh partitioning for cache-efficient visualization.

PubMed

Tchiboukdjian, Marc; Danjean, Vincent; Raffin, Bruno

2010-01-01

One important bottleneck when visualizing large data sets is the data transfer between processor and memory. Cache-aware (CA) and cache-oblivious (CO) algorithms take into consideration the memory hierarchy to design cache efficient algorithms. CO approaches have the advantage to adapt to unknown and varying memory hierarchies. Recent CA and CO algorithms developed for 3D mesh layouts significantly improve performance of previous approaches, but they lack of theoretical performance guarantees. We present in this paper a {\\schmi O}(N\\log N) algorithm to compute a CO layout for unstructured but well shaped meshes. We prove that a coherent traversal of a N-size mesh in dimension d induces less than N/B+{\\schmi O}(N/M;{1/d}) cache-misses where B and M are the block size and the cache size, respectively. Experiments show that our layout computation is faster and significantly less memory consuming than the best known CO algorithm. Performance is comparable to this algorithm for classical visualization algorithm access patterns, or better when the BSP tree produced while computing the layout is used as an acceleration data structure adjusted to the layout. We also show that cache oblivious approaches lead to significant performance increases on recent GPU architectures.

Towards reversible basic linear algebra subprograms: A performance study

DOE PAGES

Perumalla, Kalyan S.; Yoginath, Srikanth B.

2014-12-06

Problems such as fault tolerance and scalable synchronization can be efficiently solved using reversibility of applications. Making applications reversible by relying on computation rather than on memory is ideal for large scale parallel computing, especially for the next generation of supercomputers in which memory is expensive in terms of latency, energy, and price. In this direction, a case study is presented here in reversing a computational core, namely, Basic Linear Algebra Subprograms, which is widely used in scientific applications. A new Reversible BLAS (RBLAS) library interface has been designed, and a prototype has been implemented with two modes: (1) amore » memory-mode in which reversibility is obtained by checkpointing to memory in forward and restoring from memory in reverse, and (2) a computational-mode in which nothing is saved in the forward, but restoration is done entirely via inverse computation in reverse. The article is focused on detailed performance benchmarking to evaluate the runtime dynamics and performance effects, comparing reversible computation with checkpointing on both traditional CPU platforms and recent GPU accelerator platforms. For BLAS Level-1 subprograms, data indicates over an order of magnitude better speed of reversible computation compared to checkpointing. For BLAS Level-2 and Level-3, a more complex tradeoff is observed between reversible computation and checkpointing, depending on computational and memory complexities of the subprograms.« less

A communication-avoiding, hybrid-parallel, rank-revealing orthogonalization method.

DOE Office of Scientific and Technical Information (OSTI.GOV)

Hoemmen, Mark

2010-11-01

Orthogonalization consumes much of the run time of many iterative methods for solving sparse linear systems and eigenvalue problems. Commonly used algorithms, such as variants of Gram-Schmidt or Householder QR, have performance dominated by communication. Here, 'communication' includes both data movement between the CPU and memory, and messages between processors in parallel. Our Tall Skinny QR (TSQR) family of algorithms requires asymptotically fewer messages between processors and data movement between CPU and memory than typical orthogonalization methods, yet achieves the same accuracy as Householder QR factorization. Furthermore, in block orthogonalizations, TSQR is faster and more accurate than existing approaches formore » orthogonalizing the vectors within each block ('normalization'). TSQR's rank-revealing capability also makes it useful for detecting deflation in block iterative methods, for which existing approaches sacrifice performance, accuracy, or both. We have implemented a version of TSQR that exploits both distributed-memory and shared-memory parallelism, and supports real and complex arithmetic. Our implementation is optimized for the case of orthogonalizing a small number (5-20) of very long vectors. The shared-memory parallel component uses Intel's Threading Building Blocks, though its modular design supports other shared-memory programming models as well, including computation on the GPU. Our implementation achieves speedups of 2 times or more over competing orthogonalizations. It is available now in the development branch of the Trilinos software package, and will be included in the 10.8 release.« less

Accessing global data from accelerator devices

DOEpatents

Bertolli, Carlo; O'Brien, John K.; Sallenave, Olivier H.; Sura, Zehra N.

2016-12-06

An aspect includes a table of contents (TOC) that was generated by a compiler being received at an accelerator device. The TOC includes an address of global data in a host memory space. The global data is copied from the address in the host memory space to an address in the device memory space. The address in the host memory space is obtained from the received TOC. The received TOC is updated to indicate that global data is stored at the address in the device memory space. A kernel that accesses the global data from the address in the device memory space is executed. The address in the device memory space is obtained based on contents of the updated TOC. When the executing is completed, the global data from the address in the device memory space is copied to the address in the host memory space.

Accessing global data from accelerator devices

DOE Office of Scientific and Technical Information (OSTI.GOV)

Bertolli, Carlo; O'Brien, John K.; Sallenave, Olivier H.

2016-12-06

An aspect includes a table of contents (TOC) that was generated by a compiler being received at an accelerator device. The TOC includes an address of global data in a host memory space. The global data is copied from the address in the host memory space to an address in the device memory space. The address in the host memory space is obtained from the received TOC. The received TOC is updated to indicate that global data is stored at the address in the device memory space. A kernel that accesses the global data from the address in the devicemore » memory space is executed. The address in the device memory space is obtained based on contents of the updated TOC. When the executing is completed, the global data from the address in the device memory space is copied to the address in the host memory space.« less

GPU Accelerated Chemical Similarity Calculation for Compound Library Comparison

PubMed Central

Ma, Chao; Wang, Lirong; Xie, Xiang-Qun

2012-01-01

Chemical similarity calculation plays an important role in compound library design, virtual screening, and “lead” optimization. In this manuscript, we present a novel GPU-accelerated algorithm for all-vs-all Tanimoto matrix calculation and nearest neighbor search. By taking advantage of multi-core GPU architecture and CUDA parallel programming technology, the algorithm is up to 39 times superior to the existing commercial software that runs on CPUs. Because of the utilization of intrinsic GPU instructions, this approach is nearly 10 times faster than existing GPU-accelerated sparse vector algorithm, when Unity fingerprints are used for Tanimoto calculation. The GPU program that implements this new method takes about 20 minutes to complete the calculation of Tanimoto coefficients between 32M PubChem compounds and 10K Active Probes compounds, i.e., 324G Tanimoto coefficients, on a 128-CUDA-core GPU. PMID:21692447

WE-AB-303-09: Rapid Projection Computations for On-Board Digital Tomosynthesis in Radiation Therapy

DOE Office of Scientific and Technical Information (OSTI.GOV)

Iliopoulos, AS; Sun, X; Pitsianis, N

2015-06-15

Purpose: To facilitate fast and accurate iterative volumetric image reconstruction from limited-angle on-board projections. Methods: Intrafraction motion hinders the clinical applicability of modern radiotherapy techniques, such as lung stereotactic body radiation therapy (SBRT). The LIVE system may impact clinical practice by recovering volumetric information via Digital Tomosynthesis (DTS), thus entailing low time and radiation dose for image acquisition during treatment. The DTS is estimated as a deformation of prior CT via iterative registration with on-board images; this shifts the challenge to the computational domain, owing largely to repeated projection computations across iterations. We address this issue by composing efficient digitalmore » projection operators from their constituent parts. This allows us to separate the static (projection geometry) and dynamic (volume/image data) parts of projection operations by means of pre-computations, enabling fast on-board processing, while also relaxing constraints on underlying numerical models (e.g. regridding interpolation kernels). Further decoupling the projectors into simpler ones ensures the incurred memory overhead remains low, within the capacity of a single GPU. These operators depend only on the treatment plan and may be reused across iterations and patients. The dynamic processing load is kept to a minimum and maps well to the GPU computational model. Results: We have integrated efficient, pre-computable modules for volumetric ray-casting and FDK-based back-projection with the LIVE processing pipeline. Our results show a 60x acceleration of the DTS computations, compared to the previous version, using a single GPU; presently, reconstruction is attained within a couple of minutes. The present implementation allows for significant flexibility in terms of the numerical and operational projection model; we are investigating the benefit of further optimizations and accurate digital projection sub-kernels. Conclusion: Composable projection operators constitute a versatile research tool which can greatly accelerate iterative registration algorithms and may be conducive to the clinical applicability of LIVE. National Institutes of Health Grant No. R01-CA184173; GPU donation by NVIDIA Corporation.« less

GPU accelerated Monte-Carlo simulation of SEM images for metrology

NASA Astrophysics Data System (ADS)

Verduin, T.; Lokhorst, S. R.; Hagen, C. W.

2016-03-01

In this work we address the computation times of numerical studies in dimensional metrology. In particular, full Monte-Carlo simulation programs for scanning electron microscopy (SEM) image acquisition are known to be notoriously slow. Our quest in reducing the computation time of SEM image simulation has led us to investigate the use of graphics processing units (GPUs) for metrology. We have succeeded in creating a full Monte-Carlo simulation program for SEM images, which runs entirely on a GPU. The physical scattering models of this GPU simulator are identical to a previous CPU-based simulator, which includes the dielectric function model for inelastic scattering and also refinements for low-voltage SEM applications. As a case study for the performance, we considered the simulated exposure of a complex feature: an isolated silicon line with rough sidewalls located on a at silicon substrate. The surface of the rough feature is decomposed into 408 012 triangles. We have used an exposure dose of 6 mC/cm2, which corresponds to 6 553 600 primary electrons on average (Poisson distributed). We repeat the simulation for various primary electron energies, 300 eV, 500 eV, 800 eV, 1 keV, 3 keV and 5 keV. At first we run the simulation on a GeForce GTX480 from NVIDIA. The very same simulation is duplicated on our CPU-based program, for which we have used an Intel Xeon X5650. Apart from statistics in the simulation, no difference is found between the CPU and GPU simulated results. The GTX480 generates the images (depending on the primary electron energy) 350 to 425 times faster than a single threaded Intel X5650 CPU. Although this is a tremendous speedup, we actually have not reached the maximum throughput because of the limited amount of available memory on the GTX480. Nevertheless, the speedup enables the fast acquisition of simulated SEM images for metrology. We now have the potential to investigate case studies in CD-SEM metrology, which otherwise would take unreasonable amounts of computation time.

Optimizing ion channel models using a parallel genetic algorithm on graphical processors.

PubMed

Ben-Shalom, Roy; Aviv, Amit; Razon, Benjamin; Korngreen, Alon

2012-01-01

We have recently shown that we can semi-automatically constrain models of voltage-gated ion channels by combining a stochastic search algorithm with ionic currents measured using multiple voltage-clamp protocols. Although numerically successful, this approach is highly demanding computationally, with optimization on a high performance Linux cluster typically lasting several days. To solve this computational bottleneck we converted our optimization algorithm for work on a graphical processing unit (GPU) using NVIDIA's CUDA. Parallelizing the process on a Fermi graphic computing engine from NVIDIA increased the speed ∼180 times over an application running on an 80 node Linux cluster, considerably reducing simulation times. This application allows users to optimize models for ion channel kinetics on a single, inexpensive, desktop "super computer," greatly reducing the time and cost of building models relevant to neuronal physiology. We also demonstrate that the point of algorithm parallelization is crucial to its performance. We substantially reduced computing time by solving the ODEs (Ordinary Differential Equations) so as to massively reduce memory transfers to and from the GPU. This approach may be applied to speed up other data intensive applications requiring iterative solutions of ODEs. Copyright © 2012 Elsevier B.V. All rights reserved.

MicroHH 1.0: a computational fluid dynamics code for direct numerical simulation and large-eddy simulation of atmospheric boundary layer flows

NASA Astrophysics Data System (ADS)

van Heerwaarden, Chiel C.; van Stratum, Bart J. H.; Heus, Thijs; Gibbs, Jeremy A.; Fedorovich, Evgeni; Mellado, Juan Pedro

2017-08-01

This paper describes MicroHH 1.0, a new and open-source (www.microhh.org) computational fluid dynamics code for the simulation of turbulent flows in the atmosphere. It is primarily made for direct numerical simulation but also supports large-eddy simulation (LES). The paper covers the description of the governing equations, their numerical implementation, and the parameterizations included in the code. Furthermore, the paper presents the validation of the dynamical core in the form of convergence and conservation tests, and comparison of simulations of channel flows and slope flows against well-established test cases. The full numerical model, including the associated parameterizations for LES, has been tested for a set of cases under stable and unstable conditions, under the Boussinesq and anelastic approximations, and with dry and moist convection under stationary and time-varying boundary conditions. The paper presents performance tests showing good scaling from 256 to 32 768 processes. The graphical processing unit (GPU)-enabled version of the code can reach a speedup of more than an order of magnitude for simulations that fit in the memory of a single GPU.

fast_protein_cluster: parallel and optimized clustering of large-scale protein modeling data.

PubMed

Hung, Ling-Hong; Samudrala, Ram

2014-06-15

fast_protein_cluster is a fast, parallel and memory efficient package used to cluster 60 000 sets of protein models (with up to 550 000 models per set) generated by the Nutritious Rice for the World project. fast_protein_cluster is an optimized and extensible toolkit that supports Root Mean Square Deviation after optimal superposition (RMSD) and Template Modeling score (TM-score) as metrics. RMSD calculations using a laptop CPU are 60× faster than qcprot and 3× faster than current graphics processing unit (GPU) implementations. New GPU code further increases the speed of RMSD and TM-score calculations. fast_protein_cluster provides novel k-means and hierarchical clustering methods that are up to 250× and 2000× faster, respectively, than Clusco, and identify significantly more accurate models than Spicker and Clusco. fast_protein_cluster is written in C++ using OpenMP for multi-threading support. Custom streaming Single Instruction Multiple Data (SIMD) extensions and advanced vector extension intrinsics code accelerate CPU calculations, and OpenCL kernels support AMD and Nvidia GPUs. fast_protein_cluster is available under the M.I.T. license. (http://software.compbio.washington.edu/fast_protein_cluster) © The Author 2014. Published by Oxford University Press.

High-Speed Particle-in-Cell Simulation Parallelized with Graphic Processing Units for Low Temperature Plasmas for Material Processing

NASA Astrophysics Data System (ADS)

Hur, Min Young; Verboncoeur, John; Lee, Hae June

2014-10-01

Particle-in-cell (PIC) simulations have high fidelity in the plasma device requiring transient kinetic modeling compared with fluid simulations. It uses less approximation on the plasma kinetics but requires many particles and grids to observe the semantic results. It means that the simulation spends lots of simulation time in proportion to the number of particles. Therefore, PIC simulation needs high performance computing. In this research, a graphic processing unit (GPU) is adopted for high performance computing of PIC simulation for low temperature discharge plasmas. GPUs have many-core processors and high memory bandwidth compared with a central processing unit (CPU). NVIDIA GeForce GPUs were used for the test with hundreds of cores which show cost-effective performance. PIC code algorithm is divided into two modules which are a field solver and a particle mover. The particle mover module is divided into four routines which are named move, boundary, Monte Carlo collision (MCC), and deposit. Overall, the GPU code solves particle motions as well as electrostatic potential in two-dimensional geometry almost 30 times faster than a single CPU code. This work was supported by the Korea Institute of Science Technology Information.

Parallel hyperspectral image reconstruction using random projections

NASA Astrophysics Data System (ADS)

Sevilla, Jorge; Martín, Gabriel; Nascimento, José M. P.

2016-10-01

Spaceborne sensors systems are characterized by scarce onboard computing and storage resources and by communication links with reduced bandwidth. Random projections techniques have been demonstrated as an effective and very light way to reduce the number of measurements in hyperspectral data, thus, the data to be transmitted to the Earth station is reduced. However, the reconstruction of the original data from the random projections may be computationally expensive. SpeCA is a blind hyperspectral reconstruction technique that exploits the fact that hyperspectral vectors often belong to a low dimensional subspace. SpeCA has shown promising results in the task of recovering hyperspectral data from a reduced number of random measurements. In this manuscript we focus on the implementation of the SpeCA algorithm for graphics processing units (GPU) using the compute unified device architecture (CUDA). Experimental results conducted using synthetic and real hyperspectral datasets on the GPU architecture by NVIDIA: GeForce GTX 980, reveal that the use of GPUs can provide real-time reconstruction. The achieved speedup is up to 22 times when compared with the processing time of SpeCA running on one core of the Intel i7-4790K CPU (3.4GHz), with 32 Gbyte memory.

SIproc: an open-source biomedical data processing platform for large hyperspectral images.

PubMed

Berisha, Sebastian; Chang, Shengyuan; Saki, Sam; Daeinejad, Davar; He, Ziqi; Mankar, Rupali; Mayerich, David

2017-04-10

There has recently been significant interest within the vibrational spectroscopy community to apply quantitative spectroscopic imaging techniques to histology and clinical diagnosis. However, many of the proposed methods require collecting spectroscopic images that have a similar region size and resolution to the corresponding histological images. Since spectroscopic images contain significantly more spectral samples than traditional histology, the resulting data sets can approach hundreds of gigabytes to terabytes in size. This makes them difficult to store and process, and the tools available to researchers for handling large spectroscopic data sets are limited. Fundamental mathematical tools, such as MATLAB, Octave, and SciPy, are extremely powerful but require that the data be stored in fast memory. This memory limitation becomes impractical for even modestly sized histological images, which can be hundreds of gigabytes in size. In this paper, we propose an open-source toolkit designed to perform out-of-core processing of hyperspectral images. By taking advantage of graphical processing unit (GPU) computing combined with adaptive data streaming, our software alleviates common workstation memory limitations while achieving better performance than existing applications.

High Performance Implementation of 3D Convolutional Neural Networks on a GPU.

PubMed

Lan, Qiang; Wang, Zelong; Wen, Mei; Zhang, Chunyuan; Wang, Yijie

2017-01-01

Convolutional neural networks have proven to be highly successful in applications such as image classification, object tracking, and many other tasks based on 2D inputs. Recently, researchers have started to apply convolutional neural networks to video classification, which constitutes a 3D input and requires far larger amounts of memory and much more computation. FFT based methods can reduce the amount of computation, but this generally comes at the cost of an increased memory requirement. On the other hand, the Winograd Minimal Filtering Algorithm (WMFA) can reduce the number of operations required and thus can speed up the computation, without increasing the required memory. This strategy was shown to be successful for 2D neural networks. We implement the algorithm for 3D convolutional neural networks and apply it to a popular 3D convolutional neural network which is used to classify videos and compare it to cuDNN. For our highly optimized implementation of the algorithm, we observe a twofold speedup for most of the 3D convolution layers of our test network compared to the cuDNN version.

High Performance Implementation of 3D Convolutional Neural Networks on a GPU

PubMed Central

Wang, Zelong; Wen, Mei; Zhang, Chunyuan; Wang, Yijie

2017-01-01

Convolutional neural networks have proven to be highly successful in applications such as image classification, object tracking, and many other tasks based on 2D inputs. Recently, researchers have started to apply convolutional neural networks to video classification, which constitutes a 3D input and requires far larger amounts of memory and much more computation. FFT based methods can reduce the amount of computation, but this generally comes at the cost of an increased memory requirement. On the other hand, the Winograd Minimal Filtering Algorithm (WMFA) can reduce the number of operations required and thus can speed up the computation, without increasing the required memory. This strategy was shown to be successful for 2D neural networks. We implement the algorithm for 3D convolutional neural networks and apply it to a popular 3D convolutional neural network which is used to classify videos and compare it to cuDNN. For our highly optimized implementation of the algorithm, we observe a twofold speedup for most of the 3D convolution layers of our test network compared to the cuDNN version. PMID:29250109

SU-E-T-493: Accelerated Monte Carlo Methods for Photon Dosimetry Using a Dual-GPU System and CUDA.

PubMed

Liu, T; Ding, A; Xu, X

2012-06-01

To develop a Graphics Processing Unit (GPU) based Monte Carlo (MC) code that accelerates dose calculations on a dual-GPU system. We simulated a clinical case of prostate cancer treatment. A voxelized abdomen phantom derived from 120 CT slices was used containing 218×126×60 voxels, and a GE LightSpeed 16-MDCT scanner was modeled. A CPU version of the MC code was first developed in C++ and tested on Intel Xeon X5660 2.8GHz CPU, then it was translated into GPU version using CUDA C 4.1 and run on a dual Tesla m 2 090 GPU system. The code was featured with automatic assignment of simulation task to multiple GPUs, as well as accurate calculation of energy- and material- dependent cross-sections. Double-precision floating point format was used for accuracy. Doses to the rectum, prostate, bladder and femoral heads were calculated. When running on a single GPU, the MC GPU code was found to be ×19 times faster than the CPU code and ×42 times faster than MCNPX. These speedup factors were doubled on the dual-GPU system. The dose Result was benchmarked against MCNPX and a maximum difference of 1% was observed when the relative error is kept below 0.1%. A GPU-based MC code was developed for dose calculations using detailed patient and CT scanner models. Efficiency and accuracy were both guaranteed in this code. Scalability of the code was confirmed on the dual-GPU system. © 2012 American Association of Physicists in Medicine.

Changes in Brain Network Efficiency and Working Memory Performance in Aging

PubMed Central

Stanley, Matthew L.; Simpson, Sean L.; Dagenbach, Dale; Lyday, Robert G.; Burdette, Jonathan H.; Laurienti, Paul J.

2015-01-01

Working memory is a complex psychological construct referring to the temporary storage and active processing of information. We used functional connectivity brain network metrics quantifying local and global efficiency of information transfer for predicting individual variability in working memory performance on an n-back task in both young (n = 14) and older (n = 15) adults. Individual differences in both local and global efficiency during the working memory task were significant predictors of working memory performance in addition to age (and an interaction between age and global efficiency). Decreases in local efficiency during the working memory task were associated with better working memory performance in both age cohorts. In contrast, increases in global efficiency were associated with much better working performance for young participants; however, increases in global efficiency were associated with a slight decrease in working memory performance for older participants. Individual differences in local and global efficiency during resting-state sessions were not significant predictors of working memory performance. Significant group whole-brain functional network decreases in local efficiency also were observed during the working memory task compared to rest, whereas no significant differences were observed in network global efficiency. These results are discussed in relation to recently developed models of age-related differences in working memory. PMID:25875001

Changes in brain network efficiency and working memory performance in aging.

PubMed

Stanley, Matthew L; Simpson, Sean L; Dagenbach, Dale; Lyday, Robert G; Burdette, Jonathan H; Laurienti, Paul J

2015-01-01

Working memory is a complex psychological construct referring to the temporary storage and active processing of information. We used functional connectivity brain network metrics quantifying local and global efficiency of information transfer for predicting individual variability in working memory performance on an n-back task in both young (n = 14) and older (n = 15) adults. Individual differences in both local and global efficiency during the working memory task were significant predictors of working memory performance in addition to age (and an interaction between age and global efficiency). Decreases in local efficiency during the working memory task were associated with better working memory performance in both age cohorts. In contrast, increases in global efficiency were associated with much better working performance for young participants; however, increases in global efficiency were associated with a slight decrease in working memory performance for older participants. Individual differences in local and global efficiency during resting-state sessions were not significant predictors of working memory performance. Significant group whole-brain functional network decreases in local efficiency also were observed during the working memory task compared to rest, whereas no significant differences were observed in network global efficiency. These results are discussed in relation to recently developed models of age-related differences in working memory.

Rapid processing of PET list-mode data for efficient uncertainty estimation and data analysis

NASA Astrophysics Data System (ADS)

Markiewicz, P. J.; Thielemans, K.; Schott, J. M.; Atkinson, D.; Arridge, S. R.; Hutton, B. F.; Ourselin, S.

2016-07-01

In this technical note we propose a rapid and scalable software solution for the processing of PET list-mode data, which allows the efficient integration of list mode data processing into the workflow of image reconstruction and analysis. All processing is performed on the graphics processing unit (GPU), making use of streamed and concurrent kernel execution together with data transfers between disk and CPU memory as well as CPU and GPU memory. This approach leads to fast generation of multiple bootstrap realisations, and when combined with fast image reconstruction and analysis, it enables assessment of uncertainties of any image statistic and of any component of the image generation process (e.g. random correction, image processing) within reasonable time frames (e.g. within five minutes per realisation). This is of particular value when handling complex chains of image generation and processing. The software outputs the following: (1) estimate of expected random event data for noise reduction; (2) dynamic prompt and random sinograms of span-1 and span-11 and (3) variance estimates based on multiple bootstrap realisations of (1) and (2) assuming reasonable count levels for acceptable accuracy. In addition, the software produces statistics and visualisations for immediate quality control and crude motion detection, such as: (1) count rate curves; (2) centre of mass plots of the radiodistribution for motion detection; (3) video of dynamic projection views for fast visual list-mode skimming and inspection; (4) full normalisation factor sinograms. To demonstrate the software, we present an example of the above processing for fast uncertainty estimation of regional SUVR (standard uptake value ratio) calculation for a single PET scan of 18F-florbetapir using the Siemens Biograph mMR scanner.

A Real-Time Capable Software-Defined Receiver Using GPU for Adaptive Anti-Jam GPS Sensors

PubMed Central

Seo, Jiwon; Chen, Yu-Hsuan; De Lorenzo, David S.; Lo, Sherman; Enge, Per; Akos, Dennis; Lee, Jiyun

2011-01-01

Due to their weak received signal power, Global Positioning System (GPS) signals are vulnerable to radio frequency interference. Adaptive beam and null steering of the gain pattern of a GPS antenna array can significantly increase the resistance of GPS sensors to signal interference and jamming. Since adaptive array processing requires intensive computational power, beamsteering GPS receivers were usually implemented using hardware such as field-programmable gate arrays (FPGAs). However, a software implementation using general-purpose processors is much more desirable because of its flexibility and cost effectiveness. This paper presents a GPS software-defined radio (SDR) with adaptive beamsteering capability for anti-jam applications. The GPS SDR design is based on an optimized desktop parallel processing architecture using a quad-core Central Processing Unit (CPU) coupled with a new generation Graphics Processing Unit (GPU) having massively parallel processors. This GPS SDR demonstrates sufficient computational capability to support a four-element antenna array and future GPS L5 signal processing in real time. After providing the details of our design and optimization schemes for future GPU-based GPS SDR developments, the jamming resistance of our GPS SDR under synthetic wideband jamming is presented. Since the GPS SDR uses commercial-off-the-shelf hardware and processors, it can be easily adopted in civil GPS applications requiring anti-jam capabilities. PMID:22164116

A real-time capable software-defined receiver using GPU for adaptive anti-jam GPS sensors.

PubMed

Seo, Jiwon; Chen, Yu-Hsuan; De Lorenzo, David S; Lo, Sherman; Enge, Per; Akos, Dennis; Lee, Jiyun

2011-01-01

Due to their weak received signal power, Global Positioning System (GPS) signals are vulnerable to radio frequency interference. Adaptive beam and null steering of the gain pattern of a GPS antenna array can significantly increase the resistance of GPS sensors to signal interference and jamming. Since adaptive array processing requires intensive computational power, beamsteering GPS receivers were usually implemented using hardware such as field-programmable gate arrays (FPGAs). However, a software implementation using general-purpose processors is much more desirable because of its flexibility and cost effectiveness. This paper presents a GPS software-defined radio (SDR) with adaptive beamsteering capability for anti-jam applications. The GPS SDR design is based on an optimized desktop parallel processing architecture using a quad-core Central Processing Unit (CPU) coupled with a new generation Graphics Processing Unit (GPU) having massively parallel processors. This GPS SDR demonstrates sufficient computational capability to support a four-element antenna array and future GPS L5 signal processing in real time. After providing the details of our design and optimization schemes for future GPU-based GPS SDR developments, the jamming resistance of our GPS SDR under synthetic wideband jamming is presented. Since the GPS SDR uses commercial-off-the-shelf hardware and processors, it can be easily adopted in civil GPS applications requiring anti-jam capabilities.

Combating the Reliability Challenge of GPU Register File at Low Supply Voltage

DOE Office of Scientific and Technical Information (OSTI.GOV)

Tan, Jingweijia; Song, Shuaiwen; Yan, Kaige

Supply voltage reduction is an effective approach to significantly reduce GPU energy consumption. As the largest on-chip storage structure, the GPU register file becomes the reliability hotspot that prevents further supply voltage reduction below the safe limit (Vmin) due to process variation effects. This work addresses the reliability challenge of the GPU register file at low supply voltages, which is an essential first step for aggressive supply voltage reduction of the entire GPU chip. We propose GR-Guard, an architectural solution that leverages long register dead time to enable reliable operations from unreliable register file at low voltages.

A new implementation of full resolution SBAS-DInSAR processing chain for the effective monitoring of structures and infrastructures

NASA Astrophysics Data System (ADS)

Bonano, Manuela; Buonanno, Sabatino; Ojha, Chandrakanta; Berardino, Paolo; Lanari, Riccardo; Zeni, Giovanni; Manunta, Michele

2017-04-01

The advanced DInSAR technique referred to as Small BAseline Subset (SBAS) algorithm has already largely demonstrated its effectiveness to carry out multi-scale and multi-platform surface deformation analyses relevant to both natural and man-made hazards. Thanks to its capability to generate displacement maps and long-term deformation time series at both regional (low resolution analysis) and local (full resolution analysis) spatial scales, it allows to get more insights on the spatial and temporal patterns of localized displacements relevant to single buildings and infrastructures over extended urban areas, with a key role in supporting risk mitigation and preservation activities. The extensive application of the multi-scale SBAS-DInSAR approach in many scientific contexts has gone hand in hand with new SAR satellite mission development, characterized by different frequency bands, spatial resolution, revisit times and ground coverage. This brought to the generation of huge DInSAR data stacks to be efficiently handled, processed and archived, with a strong impact on both the data storage and the computational requirements needed for generating the full resolution SBAS-DInSAR results. Accordingly, innovative and effective solutions for the automatic processing of massive SAR data archives and for the operational management of the derived SBAS-DInSAR products need to be designed and implemented, by exploiting the high efficiency (in terms of portability, scalability and computing performances) of the new ICT methodologies. In this work, we present a novel parallel implementation of the full resolution SBAS-DInSAR processing chain, aimed at investigating localized displacements affecting single buildings and infrastructures relevant to very large urban areas, relying on different granularity level parallelization strategies. The image granularity level is applied in most steps of the SBAS-DInSAR processing chain and exploits the multiprocessor systems with distributed memory. Moreover, in some processing steps very heavy from the computational point of view, the Graphical Processing Units (GPU) are exploited for the processing of blocks working on a pixel-by-pixel basis, requiring strong modifications on some key parts of the sequential full resolution SBAS-DInSAR processing chain. GPU processing is implemented by efficiently exploiting parallel processing architectures (as CUDA) for increasing the computing performances, in terms of optimization of the available GPU memory, as well as reduction of the Input/Output operations on the GPU and of the whole processing time for specific blocks w.r.t. the corresponding sequential implementation, particularly critical in presence of huge DInSAR datasets. Moreover, to efficiently handle the massive amount of DInSAR measurements provided by the new generation SAR constellations (CSK and Sentinel-1), we perform a proper re-design strategy aimed at the robust assimilation of the full resolution SBAS-DInSAR results into the web-based Geonode platform of the Spatial Data Infrastructure, thus allowing the efficient management, analysis and integration of the interferometric results with different data sources.

The Performance Improvement of the Lagrangian Particle Dispersion Model (LPDM) Using Graphics Processing Unit (GPU) Computing

DTIC Science & Technology

2017-08-01

access to the GPU for general purpose processing .5 CUDA is designed to work easily with multiple programming languages , including Fortran. CUDA is a...Using Graphics Processing Unit (GPU) Computing by Leelinda P Dawson Approved for public release; distribution unlimited...The Performance Improvement of the Lagrangian Particle Dispersion Model (LPDM) Using Graphics Processing Unit (GPU) Computing by Leelinda

Validation of GPU based TomoTherapy dose calculation engine.

PubMed

Chen, Quan; Lu, Weiguo; Chen, Yu; Chen, Mingli; Henderson, Douglas; Sterpin, Edmond

2012-04-01

The graphic processing unit (GPU) based TomoTherapy convolution/superposition(C/S) dose engine (GPU dose engine) achieves a dramatic performance improvement over the traditional CPU-cluster based TomoTherapy dose engine (CPU dose engine). Besides the architecture difference between the GPU and CPU, there are several algorithm changes from the CPU dose engine to the GPU dose engine. These changes made the GPU dose slightly different from the CPU-cluster dose. In order for the commercial release of the GPU dose engine, its accuracy has to be validated. Thirty eight TomoTherapy phantom plans and 19 patient plans were calculated with both dose engines to evaluate the equivalency between the two dose engines. Gamma indices (Γ) were used for the equivalency evaluation. The GPU dose was further verified with the absolute point dose measurement with ion chamber and film measurements for phantom plans. Monte Carlo calculation was used as a reference for both dose engines in the accuracy evaluation in heterogeneous phantom and actual patients. The GPU dose engine showed excellent agreement with the current CPU dose engine. The majority of cases had over 99.99% of voxels with Γ(1%, 1 mm) < 1. The worst case observed in the phantom had 0.22% voxels violating the criterion. In patient cases, the worst percentage of voxels violating the criterion was 0.57%. For absolute point dose verification, all cases agreed with measurement to within ±3% with average error magnitude within 1%. All cases passed the acceptance criterion that more than 95% of the pixels have Γ(3%, 3 mm) < 1 in film measurement, and the average passing pixel percentage is 98.5%-99%. The GPU dose engine also showed similar degree of accuracy in heterogeneous media as the current TomoTherapy dose engine. It is verified and validated that the ultrafast TomoTherapy GPU dose engine can safely replace the existing TomoTherapy cluster based dose engine without degradation in dose accuracy.

SU-E-J-91: FFT Based Medical Image Registration Using a Graphics Processing Unit (GPU).

PubMed

Luce, J; Hoggarth, M; Lin, J; Block, A; Roeske, J

2012-06-01

To evaluate the efficiency gains obtained from using a Graphics Processing Unit (GPU) to perform a Fourier Transform (FT) based image registration. Fourier-based image registration involves obtaining the FT of the component images, and analyzing them in Fourier space to determine the translations and rotations of one image set relative to another. An important property of FT registration is that by enlarging the images (adding additional pixels), one can obtain translations and rotations with sub-pixel resolution. The expense, however, is an increased computational time. GPUs may decrease the computational time associated with FT image registration by taking advantage of their parallel architecture to perform matrix computations much more efficiently than a Central Processor Unit (CPU). In order to evaluate the computational gains produced by a GPU, images with known translational shifts were utilized. A program was written in the Interactive Data Language (IDL; Exelis, Boulder, CO) to performCPU-based calculations. Subsequently, the program was modified using GPU bindings (Tech-X, Boulder, CO) to perform GPU-based computation on the same system. Multiple image sizes were used, ranging from 256×256 to 2304×2304. The time required to complete the full algorithm by the CPU and GPU were benchmarked and the speed increase was defined as the ratio of the CPU-to-GPU computational time. The ratio of the CPU-to- GPU time was greater than 1.0 for all images, which indicates the GPU is performing the algorithm faster than the CPU. The smallest improvement, a 1.21 ratio, was found with the smallest image size of 256×256, and the largest speedup, a 4.25 ratio, was observed with the largest image size of 2304×2304. GPU programming resulted in a significant decrease in computational time associated with a FT image registration algorithm. The inclusion of the GPU may provide near real-time, sub-pixel registration capability. © 2012 American Association of Physicists in Medicine.

Simulating Cosmic Reionization and Its Observable Consequences

NASA Astrophysics Data System (ADS)

Shapiro, Paul

2017-01-01

I summarize recent progress in modelling the epoch of reionization by large- scale simulations of cosmic structure formation, radiative transfer and their interplay, which trace the ionization fronts that swept across the IGM, to predict observable signatures. Reionization by starlight from early galaxies affected their evolution, impacting reionization, itself, and imprinting the galaxies with a memory of reionization. Star formation suppression, e.g., may explain the observed underabundance of Local Group dwarfs relative to N-body predictions for Cold Dark Matter. I describe CoDa (''Cosmic Dawn''), the first fully-coupled radiation-hydrodynamical simulation of reionization and galaxy formation in the Local Universe, in a volume large enough to model reionization globally but with enough resolving power to follow all the atomic-cooling galactic halos in that volume. A 90 Mpc box was simulated from a constrained realization of primordial fluctuations, chosen to reproduce present-day features of the Local Group, including the Milky Way and M31, and the local universe beyond, including the Virgo cluster. The new RAMSES-CUDATON hybrid CPU-GPU code took 11 days to perform this simulation on the Titan supercomputer at Oak Ridge National Laboratory, with 4096-cubed N-body particles for the dark matter and 4096-cubed cells for the atomic gas and ionizing radiation.

In vivo retinal and choroidal hypoxia imaging using a novel activatable hypoxia-selective near-infrared fluorescent probe.

PubMed

Fukuda, Shinichi; Okuda, Kensuke; Kishino, Genichiro; Hoshi, Sujin; Kawano, Itsuki; Fukuda, Masahiro; Yamashita, Toshiharu; Beheregaray, Simone; Nagano, Masumi; Ohneda, Osamu; Nagasawa, Hideko; Oshika, Tetsuro

2016-12-01

Retinal hypoxia plays a crucial role in ocular neovascular diseases, such as diabetic retinopathy, retinopathy of prematurity, and retinal vascular occlusion. Fluorescein angiography is useful for identifying the hypoxia extent by detecting non-perfusion areas or neovascularization, but its ability to detect early stages of hypoxia is limited. Recently, in vivo fluorescent probes for detecting hypoxia have been developed; however, these have not been extensively applied in ophthalmology. We evaluated whether a novel donor-excited photo-induced electron transfer (d-PeT) system based on an activatable hypoxia-selective near-infrared fluorescent (NIRF) probe (GPU-327) responds to both mild and severe hypoxia in various ocular ischemic diseases animal models. The ocular fundus examination offers unique opportunities for direct observation of the retina through the transparent cornea and lens. After injection of GPU-327 in various ocular hypoxic diseases of mouse and rabbit models, NIRF imaging in the ocular fundus can be performed noninvasively and easily by using commercially available fundus cameras. To investigate the safety of GPU-327, electroretinograms were also recorded after GPU-327 and PBS injection. Fluorescence of GPU-327 increased under mild hypoxic conditions in vitro. GPU-327 also yielded excellent signal-to-noise ratio without washing out in vivo experiments. By using near-infrared region, GPU-327 enables imaging of deeper ischemia, such as choroidal circulation. Additionally, from an electroretinogram, GPU-327 did not cause neurotoxicity. GPU-327 identified hypoxic area both in vivo and in vitro.

Large-scale neural circuit mapping data analysis accelerated with the graphical processing unit (GPU).

PubMed

Shi, Yulin; Veidenbaum, Alexander V; Nicolau, Alex; Xu, Xiangmin

2015-01-15

Modern neuroscience research demands computing power. Neural circuit mapping studies such as those using laser scanning photostimulation (LSPS) produce large amounts of data and require intensive computation for post hoc processing and analysis. Here we report on the design and implementation of a cost-effective desktop computer system for accelerated experimental data processing with recent GPU computing technology. A new version of Matlab software with GPU enabled functions is used to develop programs that run on Nvidia GPUs to harness their parallel computing power. We evaluated both the central processing unit (CPU) and GPU-enabled computational performance of our system in benchmark testing and practical applications. The experimental results show that the GPU-CPU co-processing of simulated data and actual LSPS experimental data clearly outperformed the multi-core CPU with up to a 22× speedup, depending on computational tasks. Further, we present a comparison of numerical accuracy between GPU and CPU computation to verify the precision of GPU computation. In addition, we show how GPUs can be effectively adapted to improve the performance of commercial image processing software such as Adobe Photoshop. To our best knowledge, this is the first demonstration of GPU application in neural circuit mapping and electrophysiology-based data processing. Together, GPU enabled computation enhances our ability to process large-scale data sets derived from neural circuit mapping studies, allowing for increased processing speeds while retaining data precision. Copyright © 2014 Elsevier B.V. All rights reserved.

Large scale neural circuit mapping data analysis accelerated with the graphical processing unit (GPU)

PubMed Central

Shi, Yulin; Veidenbaum, Alexander V.; Nicolau, Alex; Xu, Xiangmin

2014-01-01

Background Modern neuroscience research demands computing power. Neural circuit mapping studies such as those using laser scanning photostimulation (LSPS) produce large amounts of data and require intensive computation for post-hoc processing and analysis. New Method Here we report on the design and implementation of a cost-effective desktop computer system for accelerated experimental data processing with recent GPU computing technology. A new version of Matlab software with GPU enabled functions is used to develop programs that run on Nvidia GPUs to harness their parallel computing power. Results We evaluated both the central processing unit (CPU) and GPU-enabled computational performance of our system in benchmark testing and practical applications. The experimental results show that the GPU-CPU co-processing of simulated data and actual LSPS experimental data clearly outperformed the multi-core CPU with up to a 22x speedup, depending on computational tasks. Further, we present a comparison of numerical accuracy between GPU and CPU computation to verify the precision of GPU computation. In addition, we show how GPUs can be effectively adapted to improve the performance of commercial image processing software such as Adobe Photoshop. Comparison with Existing Method(s) To our best knowledge, this is the first demonstration of GPU application in neural circuit mapping and electrophysiology-based data processing. Conclusions Together, GPU enabled computation enhances our ability to process large-scale data sets derived from neural circuit mapping studies, allowing for increased processing speeds while retaining data precision. PMID:25277633

Real-time image processing for non-contact monitoring of dynamic displacements using smartphone technologies

NASA Astrophysics Data System (ADS)

Min, Jae-Hong; Gelo, Nikolas J.; Jo, Hongki

2016-04-01

The newly developed smartphone application, named RINO, in this study allows measuring absolute dynamic displacements and processing them in real time using state-of-the-art smartphone technologies, such as high-performance graphics processing unit (GPU), in addition to already powerful CPU and memories, embedded high-speed/ resolution camera, and open-source computer vision libraries. A carefully designed color-patterned target and user-adjustable crop filter enable accurate and fast image processing, allowing up to 240fps for complete displacement calculation and real-time display. The performances of the developed smartphone application are experimentally validated, showing comparable accuracy with those of conventional laser displacement sensor.

GPU MrBayes V3.1: MrBayes on Graphics Processing Units for Protein Sequence Data.

PubMed

Pang, Shuai; Stones, Rebecca J; Ren, Ming-Ming; Liu, Xiao-Guang; Wang, Gang; Xia, Hong-ju; Wu, Hao-Yang; Liu, Yang; Xie, Qiang

2015-09-01

We present a modified GPU (graphics processing unit) version of MrBayes, called ta(MC)(3) (GPU MrBayes V3.1), for Bayesian phylogenetic inference on protein data sets. Our main contributions are 1) utilizing 64-bit variables, thereby enabling ta(MC)(3) to process larger data sets than MrBayes; and 2) to use Kahan summation to improve accuracy, convergence rates, and consequently runtime. Versus the current fastest software, we achieve a speedup of up to around 2.5 (and up to around 90 vs. serial MrBayes), and more on multi-GPU hardware. GPU MrBayes V3.1 is available from http://sourceforge.net/projects/mrbayes-gpu/. © The Author 2015. Published by Oxford University Press on behalf of the Society for Molecular Biology and Evolution. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.

GPU-based High-Performance Computing for Radiation Therapy

PubMed Central

Jia, Xun; Ziegenhein, Peter; Jiang, Steve B.

2014-01-01

Recent developments in radiotherapy therapy demand high computation powers to solve challenging problems in a timely fashion in a clinical environment. Graphics processing unit (GPU), as an emerging high-performance computing platform, has been introduced to radiotherapy. It is particularly attractive due to its high computational power, small size, and low cost for facility deployment and maintenance. Over the past a few years, GPU-based high-performance computing in radiotherapy has experienced rapid developments. A tremendous amount of studies have been conducted, in which large acceleration factors compared with the conventional CPU platform have been observed. In this article, we will first give a brief introduction to the GPU hardware structure and programming model. We will then review the current applications of GPU in major imaging-related and therapy-related problems encountered in radiotherapy. A comparison of GPU with other platforms will also be presented. PMID:24486639

GPU accelerated implementation of NCI calculations using promolecular density.

PubMed

Rubez, Gaëtan; Etancelin, Jean-Matthieu; Vigouroux, Xavier; Krajecki, Michael; Boisson, Jean-Charles; Hénon, Eric

2017-05-30

The NCI approach is a modern tool to reveal chemical noncovalent interactions. It is particularly attractive to describe ligand-protein binding. A custom implementation for NCI using promolecular density is presented. It is designed to leverage the computational power of NVIDIA graphics processing unit (GPU) accelerators through the CUDA programming model. The code performances of three versions are examined on a test set of 144 systems. NCI calculations are particularly well suited to the GPU architecture, which reduces drastically the computational time. On a single compute node, the dual-GPU version leads to a 39-fold improvement for the biggest instance compared to the optimal OpenMP parallel run (C code, icc compiler) with 16 CPU cores. Energy consumption measurements carried out on both CPU and GPU NCI tests show that the GPU approach provides substantial energy savings. © 2017 Wiley Periodicals, Inc. © 2017 Wiley Periodicals, Inc.

A GPU-based calculation using the three-dimensional FDTD method for electromagnetic field analysis.

PubMed

Nagaoka, Tomoaki; Watanabe, Soichi

2010-01-01

Numerical simulations with the numerical human model using the finite-difference time domain (FDTD) method have recently been performed frequently in a number of fields in biomedical engineering. However, the FDTD calculation runs too slowly. We focus, therefore, on general purpose programming on the graphics processing unit (GPGPU). The three-dimensional FDTD method was implemented on the GPU using Compute Unified Device Architecture (CUDA). In this study, we used the NVIDIA Tesla C1060 as a GPGPU board. The performance of the GPU is evaluated in comparison with the performance of a conventional CPU and a vector supercomputer. The results indicate that three-dimensional FDTD calculations using a GPU can significantly reduce run time in comparison with that using a conventional CPU, even a native GPU implementation of the three-dimensional FDTD method, while the GPU/CPU speed ratio varies with the calculation domain and thread block size.

GPU computing in medical physics: a review.

PubMed

Pratx, Guillem; Xing, Lei

2011-05-01

The graphics processing unit (GPU) has emerged as a competitive platform for computing massively parallel problems. Many computing applications in medical physics can be formulated as data-parallel tasks that exploit the capabilities of the GPU for reducing processing times. The authors review the basic principles of GPU computing as well as the main performance optimization techniques, and survey existing applications in three areas of medical physics, namely image reconstruction, dose calculation and treatment plan optimization, and image processing.

CPU-GPU hybrid accelerating the Zuker algorithm for RNA secondary structure prediction applications.

PubMed

Lei, Guoqing; Dou, Yong; Wan, Wen; Xia, Fei; Li, Rongchun; Ma, Meng; Zou, Dan

2012-01-01

Prediction of ribonucleic acid (RNA) secondary structure remains one of the most important research areas in bioinformatics. The Zuker algorithm is one of the most popular methods of free energy minimization for RNA secondary structure prediction. Thus far, few studies have been reported on the acceleration of the Zuker algorithm on general-purpose processors or on extra accelerators such as Field Programmable Gate-Array (FPGA) and Graphics Processing Units (GPU). To the best of our knowledge, no implementation combines both CPU and extra accelerators, such as GPUs, to accelerate the Zuker algorithm applications. In this paper, a CPU-GPU hybrid computing system that accelerates Zuker algorithm applications for RNA secondary structure prediction is proposed. The computing tasks are allocated between CPU and GPU for parallel cooperate execution. Performance differences between the CPU and the GPU in the task-allocation scheme are considered to obtain workload balance. To improve the hybrid system performance, the Zuker algorithm is optimally implemented with special methods for CPU and GPU architecture. Speedup of 15.93× over optimized multi-core SIMD CPU implementation and performance advantage of 16% over optimized GPU implementation are shown in the experimental results. More than 14% of the sequences are executed on CPU in the hybrid system. The system combining CPU and GPU to accelerate the Zuker algorithm is proven to be promising and can be applied to other bioinformatics applications.

Efficient Implementation of MrBayes on Multi-GPU

PubMed Central

Zhou, Jianfu; Liu, Xiaoguang; Wang, Gang

2013-01-01

MrBayes, using Metropolis-coupled Markov chain Monte Carlo (MCMCMC or (MC)3), is a popular program for Bayesian inference. As a leading method of using DNA data to infer phylogeny, the (MC)3 Bayesian algorithm and its improved and parallel versions are now not fast enough for biologists to analyze massive real-world DNA data. Recently, graphics processor unit (GPU) has shown its power as a coprocessor (or rather, an accelerator) in many fields. This article describes an efficient implementation a(MC)3 (aMCMCMC) for MrBayes (MC)3 on compute unified device architecture. By dynamically adjusting the task granularity to adapt to input data size and hardware configuration, it makes full use of GPU cores with different data sets. An adaptive method is also developed to split and combine DNA sequences to make full use of a large number of GPU cards. Furthermore, a new “node-by-node” task scheduling strategy is developed to improve concurrency, and several optimizing methods are used to reduce extra overhead. Experimental results show that a(MC)3 achieves up to 63× speedup over serial MrBayes on a single machine with one GPU card, and up to 170× speedup with four GPU cards, and up to 478× speedup with a 32-node GPU cluster. a(MC)3 is dramatically faster than all the previous (MC)3 algorithms and scales well to large GPU clusters. PMID:23493260

Efficient implementation of MrBayes on multi-GPU.

PubMed

Bao, Jie; Xia, Hongju; Zhou, Jianfu; Liu, Xiaoguang; Wang, Gang

2013-06-01

MrBayes, using Metropolis-coupled Markov chain Monte Carlo (MCMCMC or (MC)(3)), is a popular program for Bayesian inference. As a leading method of using DNA data to infer phylogeny, the (MC)(3) Bayesian algorithm and its improved and parallel versions are now not fast enough for biologists to analyze massive real-world DNA data. Recently, graphics processor unit (GPU) has shown its power as a coprocessor (or rather, an accelerator) in many fields. This article describes an efficient implementation a(MC)(3) (aMCMCMC) for MrBayes (MC)(3) on compute unified device architecture. By dynamically adjusting the task granularity to adapt to input data size and hardware configuration, it makes full use of GPU cores with different data sets. An adaptive method is also developed to split and combine DNA sequences to make full use of a large number of GPU cards. Furthermore, a new "node-by-node" task scheduling strategy is developed to improve concurrency, and several optimizing methods are used to reduce extra overhead. Experimental results show that a(MC)(3) achieves up to 63× speedup over serial MrBayes on a single machine with one GPU card, and up to 170× speedup with four GPU cards, and up to 478× speedup with a 32-node GPU cluster. a(MC)(3) is dramatically faster than all the previous (MC)(3) algorithms and scales well to large GPU clusters.

GPU-accelerated non-uniform fast Fourier transform-based compressive sensing spectral domain optical coherence tomography.

PubMed

Xu, Daguang; Huang, Yong; Kang, Jin U

2014-06-16

We implemented the graphics processing unit (GPU) accelerated compressive sensing (CS) non-uniform in k-space spectral domain optical coherence tomography (SD OCT). Kaiser-Bessel (KB) function and Gaussian function are used independently as the convolution kernel in the gridding-based non-uniform fast Fourier transform (NUFFT) algorithm with different oversampling ratios and kernel widths. Our implementation is compared with the GPU-accelerated modified non-uniform discrete Fourier transform (MNUDFT) matrix-based CS SD OCT and the GPU-accelerated fast Fourier transform (FFT)-based CS SD OCT. It was found that our implementation has comparable performance to the GPU-accelerated MNUDFT-based CS SD OCT in terms of image quality while providing more than 5 times speed enhancement. When compared to the GPU-accelerated FFT based-CS SD OCT, it shows smaller background noise and less side lobes while eliminating the need for the cumbersome k-space grid filling and the k-linear calibration procedure. Finally, we demonstrated that by using a conventional desktop computer architecture having three GPUs, real-time B-mode imaging can be obtained in excess of 30 fps for the GPU-accelerated NUFFT based CS SD OCT with frame size 2048(axial) × 1,000(lateral).

GPU-Meta-Storms: computing the structure similarities among massive amount of microbial community samples using GPU.

PubMed

Su, Xiaoquan; Wang, Xuetao; Jing, Gongchao; Ning, Kang

2014-04-01

The number of microbial community samples is increasing with exponential speed. Data-mining among microbial community samples could facilitate the discovery of valuable biological information that is still hidden in the massive data. However, current methods for the comparison among microbial communities are limited by their ability to process large amount of samples each with complex community structure. We have developed an optimized GPU-based software, GPU-Meta-Storms, to efficiently measure the quantitative phylogenetic similarity among massive amount of microbial community samples. Our results have shown that GPU-Meta-Storms would be able to compute the pair-wise similarity scores for 10 240 samples within 20 min, which gained a speed-up of >17 000 times compared with single-core CPU, and >2600 times compared with 16-core CPU. Therefore, the high-performance of GPU-Meta-Storms could facilitate in-depth data mining among massive microbial community samples, and make the real-time analysis and monitoring of temporal or conditional changes for microbial communities possible. GPU-Meta-Storms is implemented by CUDA (Compute Unified Device Architecture) and C++. Source code is available at http://www.computationalbioenergy.org/meta-storms.html.

Compute-unified device architecture implementation of a block-matching algorithm for multiple graphical processing unit cards

PubMed Central

Massanes, Francesc; Cadennes, Marie; Brankov, Jovan G.

2012-01-01

In this paper we describe and evaluate a fast implementation of a classical block matching motion estimation algorithm for multiple Graphical Processing Units (GPUs) using the Compute Unified Device Architecture (CUDA) computing engine. The implemented block matching algorithm (BMA) uses summed absolute difference (SAD) error criterion and full grid search (FS) for finding optimal block displacement. In this evaluation we compared the execution time of a GPU and CPU implementation for images of various sizes, using integer and non-integer search grids. The results show that use of a GPU card can shorten computation time by a factor of 200 times for integer and 1000 times for a non-integer search grid. The additional speedup for non-integer search grid comes from the fact that GPU has built-in hardware for image interpolation. Further, when using multiple GPU cards, the presented evaluation shows the importance of the data splitting method across multiple cards, but an almost linear speedup with a number of cards is achievable. In addition we compared execution time of the proposed FS GPU implementation with two existing, highly optimized non-full grid search CPU based motion estimations methods, namely implementation of the Pyramidal Lucas Kanade Optical flow algorithm in OpenCV and Simplified Unsymmetrical multi-Hexagon search in H.264/AVC standard. In these comparisons, FS GPU implementation still showed modest improvement even though the computational complexity of FS GPU implementation is substantially higher than non-FS CPU implementation. We also demonstrated that for an image sequence of 720×480 pixels in resolution, commonly used in video surveillance, the proposed GPU implementation is sufficiently fast for real-time motion estimation at 30 frames-per-second using two NVIDIA C1060 Tesla GPU cards. PMID:22347787

Compute-unified device architecture implementation of a block-matching algorithm for multiple graphical processing unit cards.

PubMed

Massanes, Francesc; Cadennes, Marie; Brankov, Jovan G

2011-07-01

In this paper we describe and evaluate a fast implementation of a classical block matching motion estimation algorithm for multiple Graphical Processing Units (GPUs) using the Compute Unified Device Architecture (CUDA) computing engine. The implemented block matching algorithm (BMA) uses summed absolute difference (SAD) error criterion and full grid search (FS) for finding optimal block displacement. In this evaluation we compared the execution time of a GPU and CPU implementation for images of various sizes, using integer and non-integer search grids.The results show that use of a GPU card can shorten computation time by a factor of 200 times for integer and 1000 times for a non-integer search grid. The additional speedup for non-integer search grid comes from the fact that GPU has built-in hardware for image interpolation. Further, when using multiple GPU cards, the presented evaluation shows the importance of the data splitting method across multiple cards, but an almost linear speedup with a number of cards is achievable.In addition we compared execution time of the proposed FS GPU implementation with two existing, highly optimized non-full grid search CPU based motion estimations methods, namely implementation of the Pyramidal Lucas Kanade Optical flow algorithm in OpenCV and Simplified Unsymmetrical multi-Hexagon search in H.264/AVC standard. In these comparisons, FS GPU implementation still showed modest improvement even though the computational complexity of FS GPU implementation is substantially higher than non-FS CPU implementation.We also demonstrated that for an image sequence of 720×480 pixels in resolution, commonly used in video surveillance, the proposed GPU implementation is sufficiently fast for real-time motion estimation at 30 frames-per-second using two NVIDIA C1060 Tesla GPU cards.

Study on efficiency of time computation in x-ray imaging simulation base on Monte Carlo algorithm using graphics processing unit

DOE Office of Scientific and Technical Information (OSTI.GOV)

Setiani, Tia Dwi, E-mail: tiadwisetiani@gmail.com; Suprijadi; Nuclear Physics and Biophysics Reaserch Division, Faculty of Mathematics and Natural Sciences, Institut Teknologi Bandung Jalan Ganesha 10 Bandung, 40132

Monte Carlo (MC) is one of the powerful techniques for simulation in x-ray imaging. MC method can simulate the radiation transport within matter with high accuracy and provides a natural way to simulate radiation transport in complex systems. One of the codes based on MC algorithm that are widely used for radiographic images simulation is MC-GPU, a codes developed by Andrea Basal. This study was aimed to investigate the time computation of x-ray imaging simulation in GPU (Graphics Processing Unit) compared to a standard CPU (Central Processing Unit). Furthermore, the effect of physical parameters to the quality of radiographic imagesmore » and the comparison of image quality resulted from simulation in the GPU and CPU are evaluated in this paper. The simulations were run in CPU which was simulated in serial condition, and in two GPU with 384 cores and 2304 cores. In simulation using GPU, each cores calculates one photon, so, a large number of photon were calculated simultaneously. Results show that the time simulations on GPU were significantly accelerated compared to CPU. The simulations on the 2304 core of GPU were performed about 64 -114 times faster than on CPU, while the simulation on the 384 core of GPU were performed about 20 – 31 times faster than in a single core of CPU. Another result shows that optimum quality of images from the simulation was gained at the history start from 10{sup 8} and the energy from 60 Kev to 90 Kev. Analyzed by statistical approach, the quality of GPU and CPU images are relatively the same.« less

Hadoop-MCC: Efficient Multiple Compound Comparison Algorithm Using Hadoop.

PubMed

Hua, Guan-Jie; Hung, Che-Lun; Tang, Chuan Yi

2018-01-01

In the past decade, the drug design technologies have been improved enormously. The computer-aided drug design (CADD) has played an important role in analysis and prediction in drug development, which makes the procedure more economical and efficient. However, computation with big data, such as ZINC containing more than 60 million compounds data and GDB-13 with more than 930 million small molecules, is a noticeable issue of time-consuming problem. Therefore, we propose a novel heterogeneous high performance computing method, named as Hadoop-MCC, integrating Hadoop and GPU, to copy with big chemical structure data efficiently. Hadoop-MCC gains the high availability and fault tolerance from Hadoop, as Hadoop is used to scatter input data to GPU devices and gather the results from GPU devices. Hadoop framework adopts mapper/reducer computation model. In the proposed method, mappers response for fetching SMILES data segments and perform LINGO method on GPU, then reducers collect all comparison results produced by mappers. Due to the high availability of Hadoop, all of LINGO computational jobs on mappers can be completed, even if some of the mappers encounter problems. A comparison of LINGO is performed on each the GPU device in parallel. According to the experimental results, the proposed method on multiple GPU devices can achieve better computational performance than the CUDA-MCC on a single GPU device. Hadoop-MCC is able to achieve scalability, high availability, and fault tolerance granted by Hadoop, and high performance as well by integrating computational power of both of Hadoop and GPU. It has been shown that using the heterogeneous architecture as Hadoop-MCC effectively can enhance better computational performance than on a single GPU device. Copyright© Bentham Science Publishers; For any queries, please email at epub@benthamscience.org.

GPU Computing in Bayesian Inference of Realized Stochastic Volatility Model

NASA Astrophysics Data System (ADS)

Takaishi, Tetsuya

2015-01-01

The realized stochastic volatility (RSV) model that utilizes the realized volatility as additional information has been proposed to infer volatility of financial time series. We consider the Bayesian inference of the RSV model by the Hybrid Monte Carlo (HMC) algorithm. The HMC algorithm can be parallelized and thus performed on the GPU for speedup. The GPU code is developed with CUDA Fortran. We compare the computational time in performing the HMC algorithm on GPU (GTX 760) and CPU (Intel i7-4770 3.4GHz) and find that the GPU can be up to 17 times faster than the CPU. We also code the program with OpenACC and find that appropriate coding can achieve the similar speedup with CUDA Fortran.

HASEonGPU-An adaptive, load-balanced MPI/GPU-code for calculating the amplified spontaneous emission in high power laser media

NASA Astrophysics Data System (ADS)

Eckert, C. H. J.; Zenker, E.; Bussmann, M.; Albach, D.

2016-10-01

We present an adaptive Monte Carlo algorithm for computing the amplified spontaneous emission (ASE) flux in laser gain media pumped by pulsed lasers. With the design of high power lasers in mind, which require large size gain media, we have developed the open source code HASEonGPU that is capable of utilizing multiple graphic processing units (GPUs). With HASEonGPU, time to solution is reduced to minutes on a medium size GPU cluster of 64 NVIDIA Tesla K20m GPUs and excellent speedup is achieved when scaling to multiple GPUs. Comparison of simulation results to measurements of ASE in Y b 3 + : Y AG ceramics show perfect agreement.

Local Alignment Tool Based on Hadoop Framework and GPU Architecture

PubMed Central

Hung, Che-Lun; Hua, Guan-Jie

2014-01-01

With the rapid growth of next generation sequencing technologies, such as Slex, more and more data have been discovered and published. To analyze such huge data the computational performance is an important issue. Recently, many tools, such as SOAP, have been implemented on Hadoop and GPU parallel computing architectures. BLASTP is an important tool, implemented on GPU architectures, for biologists to compare protein sequences. To deal with the big biology data, it is hard to rely on single GPU. Therefore, we implement a distributed BLASTP by combining Hadoop and multi-GPUs. The experimental results present that the proposed method can improve the performance of BLASTP on single GPU, and also it can achieve high availability and fault tolerance. PMID:24955362

Local alignment tool based on Hadoop framework and GPU architecture.

PubMed

Hung, Che-Lun; Hua, Guan-Jie

2014-01-01

With the rapid growth of next generation sequencing technologies, such as Slex, more and more data have been discovered and published. To analyze such huge data the computational performance is an important issue. Recently, many tools, such as SOAP, have been implemented on Hadoop and GPU parallel computing architectures. BLASTP is an important tool, implemented on GPU architectures, for biologists to compare protein sequences. To deal with the big biology data, it is hard to rely on single GPU. Therefore, we implement a distributed BLASTP by combining Hadoop and multi-GPUs. The experimental results present that the proposed method can improve the performance of BLASTP on single GPU, and also it can achieve high availability and fault tolerance.

Parallel Agent-Based Simulations on Clusters of GPUs and Multi-Core Processors

DOE Office of Scientific and Technical Information (OSTI.GOV)

Aaby, Brandon G; Perumalla, Kalyan S; Seal, Sudip K

2010-01-01

An effective latency-hiding mechanism is presented in the parallelization of agent-based model simulations (ABMS) with millions of agents. The mechanism is designed to accommodate the hierarchical organization as well as heterogeneity of current state-of-the-art parallel computing platforms. We use it to explore the computation vs. communication trade-off continuum available with the deep computational and memory hierarchies of extant platforms and present a novel analytical model of the tradeoff. We describe our implementation and report preliminary performance results on two distinct parallel platforms suitable for ABMS: CUDA threads on multiple, networked graphical processing units (GPUs), and pthreads on multi-core processors. Messagemore » Passing Interface (MPI) is used for inter-GPU as well as inter-socket communication on a cluster of multiple GPUs and multi-core processors. Results indicate the benefits of our latency-hiding scheme, delivering as much as over 100-fold improvement in runtime for certain benchmark ABMS application scenarios with several million agents. This speed improvement is obtained on our system that is already two to three orders of magnitude faster on one GPU than an equivalent CPU-based execution in a popular simulator in Java. Thus, the overall execution of our current work is over four orders of magnitude faster when executed on multiple GPUs.« less

FastMag: Fast micromagnetic simulator for complex magnetic structures (invited)

NASA Astrophysics Data System (ADS)

Chang, R.; Li, S.; Lubarda, M. V.; Livshitz, B.; Lomakin, V.

2011-04-01

A fast micromagnetic simulator (FastMag) for general problems is presented. FastMag solves the Landau-Lifshitz-Gilbert equation and can handle multiscale problems with a high computational efficiency. The simulator derives its high performance from efficient methods for evaluating the effective field and from implementations on massively parallel graphics processing unit (GPU) architectures. FastMag discretizes the computational domain into tetrahedral elements and therefore is highly flexible for general problems. The magnetostatic field is computed via the superposition principle for both volume and surface parts of the computational domain. This is accomplished by implementing efficient quadrature rules and analytical integration for overlapping elements in which the integral kernel is singular. Thus, discretized superposition integrals are computed using a nonuniform grid interpolation method, which evaluates the field from N sources at N collocated observers in O(N) operations. This approach allows handling objects of arbitrary shape, allows easily calculating of the field outside the magnetized domains, does not require solving a linear system of equations, and requires little memory. FastMag is implemented on GPUs with ?> GPU-central processing unit speed-ups of 2 orders of magnitude. Simulations are shown of a large array of magnetic dots and a recording head fully discretized down to the exchange length, with over a hundred million tetrahedral elements on an inexpensive desktop computer.

Transform coding for hardware-accelerated volume rendering.

PubMed

Fout, Nathaniel; Ma, Kwan-Liu

2007-01-01

Hardware-accelerated volume rendering using the GPU is now the standard approach for real-time volume rendering, although limited graphics memory can present a problem when rendering large volume data sets. Volumetric compression in which the decompression is coupled to rendering has been shown to be an effective solution to this problem; however, most existing techniques were developed in the context of software volume rendering, and all but the simplest approaches are prohibitive in a real-time hardware-accelerated volume rendering context. In this paper we present a novel block-based transform coding scheme designed specifically with real-time volume rendering in mind, such that the decompression is fast without sacrificing compression quality. This is made possible by consolidating the inverse transform with dequantization in such a way as to allow most of the reprojection to be precomputed. Furthermore, we take advantage of the freedom afforded by off-line compression in order to optimize the encoding as much as possible while hiding this complexity from the decoder. In this context we develop a new block classification scheme which allows us to preserve perceptually important features in the compression. The result of this work is an asymmetric transform coding scheme that allows very large volumes to be compressed and then decompressed in real-time while rendering on the GPU.

New algorithm for tensor contractions on multi-core CPUs, GPUs, and accelerators enables CCSD and EOM-CCSD calculations with over 1000 basis functions on a single compute node.

PubMed

Kaliman, Ilya A; Krylov, Anna I

2017-04-30

A new hardware-agnostic contraction algorithm for tensors of arbitrary symmetry and sparsity is presented. The algorithm is implemented as a stand-alone open-source code libxm. This code is also integrated with general tensor library libtensor and with the Q-Chem quantum-chemistry package. An overview of the algorithm, its implementation, and benchmarks are presented. Similarly to other tensor software, the algorithm exploits efficient matrix multiplication libraries and assumes that tensors are stored in a block-tensor form. The distinguishing features of the algorithm are: (i) efficient repackaging of the individual blocks into large matrices and back, which affords efficient graphics processing unit (GPU)-enabled calculations without modifications of higher-level codes; (ii) fully asynchronous data transfer between disk storage and fast memory. The algorithm enables canonical all-electron coupled-cluster and equation-of-motion coupled-cluster calculations with single and double substitutions (CCSD and EOM-CCSD) with over 1000 basis functions on a single quad-GPU machine. We show that the algorithm exhibits predicted theoretical scaling for canonical CCSD calculations, O(N 6 ), irrespective of the data size on disk. © 2017 Wiley Periodicals, Inc. © 2017 Wiley Periodicals, Inc.

Reconfigurable PCI Express cards for low-latency data transport in HEP experiments

NASA Astrophysics Data System (ADS)

Ammendola, R.; Biagioni, A.; Cretaro, P.; Frezza, O.; Lamanna, G.; Lo Cicero, F.; Lonardo, A.; Martinelli, M.; Paolucci, P. S.; Pastorelli, E.; Pontisso, L.; Simula, F.; Vicini, P.

2017-01-01

State-of-the-art technology supports the High Energy Physics community in addressing the problem of managing an overwhelming amount of experimental data. From the point of view of communication between the detectors' readout system and computing nodes, the critical issues are the following: latency, moving data in a deterministic and low amount of time; bandwidth, guaranteeing the maximum capability of the link and communication protocol adopted; endpoint consolidation, tight aggregation of channels on a single board. This contribution describes the status and performances of the NaNet project, whose goal is the design of a family of FPGA-based PCIe network interface cards. The efforts of the team are focused on implementing a low-latency, real-time data transport mechanism between the board network multi-channel system and CPU and GPU accelerators memories on the host. Several opportunities concerning technical solutions and scientific applications have been explored: NaNet-1 with a single GbE I/O interface, and NaNet-10, offering four 10GbE ports, for activities related to the GPU-based real-time trigger of NA62 experiment at CERN; NaNet ^3 , with four 2.5Gbit optical channels, developed for the KM3NeT-ITALIA underwater neutrino telescope.

Phase quality map based on local multi-unwrapped results for two-dimensional phase unwrapping.

PubMed

Zhong, Heping; Tang, Jinsong; Zhang, Sen

2015-02-01

The efficiency of a phase unwrapping algorithm and the reliability of the corresponding unwrapped result are two key problems in reconstructing the digital elevation model of a scene from its interferometric synthetic aperture radar (InSAR) or interferometric synthetic aperture sonar (InSAS) data. In this paper, a new phase quality map is designed and implemented in a graphic processing unit (GPU) environment, which greatly accelerates the unwrapping process of the quality-guided algorithm and enhances the correctness of the unwrapped result. In a local wrapped phase window, the center point is selected as the reference point, and then two unwrapped results are computed by integrating in two different simple ways. After the two local unwrapped results are computed, the total difference of the two unwrapped results is regarded as the phase quality value of the center point. In order to accelerate the computing process of the new proposed quality map, we have implemented it in a GPU environment. The wrapped phase data are first uploaded to the memory of a device, and then the kernel function is called in the device to compute the phase quality in parallel by blocks of threads. Unwrapping tests performed on the simulated and real InSAS data confirm the accuracy and efficiency of the proposed method.

Fast CPU-based Monte Carlo simulation for radiotherapy dose calculation.

PubMed

Ziegenhein, Peter; Pirner, Sven; Ph Kamerling, Cornelis; Oelfke, Uwe

2015-08-07

Monte-Carlo (MC) simulations are considered to be the most accurate method for calculating dose distributions in radiotherapy. Its clinical application, however, still is limited by the long runtimes conventional implementations of MC algorithms require to deliver sufficiently accurate results on high resolution imaging data. In order to overcome this obstacle we developed the software-package PhiMC, which is capable of computing precise dose distributions in a sub-minute time-frame by leveraging the potential of modern many- and multi-core CPU-based computers. PhiMC is based on the well verified dose planning method (DPM). We could demonstrate that PhiMC delivers dose distributions which are in excellent agreement to DPM. The multi-core implementation of PhiMC scales well between different computer architectures and achieves a speed-up of up to 37[Formula: see text] compared to the original DPM code executed on a modern system. Furthermore, we could show that our CPU-based implementation on a modern workstation is between 1.25[Formula: see text] and 1.95[Formula: see text] faster than a well-known GPU implementation of the same simulation method on a NVIDIA Tesla C2050. Since CPUs work on several hundreds of GB RAM the typical GPU memory limitation does not apply for our implementation and high resolution clinical plans can be calculated.

A survey of CPU-GPU heterogeneous computing techniques

DOE PAGES

Mittal, Sparsh; Vetter, Jeffrey S.

2015-07-04

As both CPU and GPU become employed in a wide range of applications, it has been acknowledged that both of these processing units (PUs) have their unique features and strengths and hence, CPU-GPU collaboration is inevitable to achieve high-performance computing. This has motivated significant amount of research on heterogeneous computing techniques, along with the design of CPU-GPU fused chips and petascale heterogeneous supercomputers. In this paper, we survey heterogeneous computing techniques (HCTs) such as workload-partitioning which enable utilizing both CPU and GPU to improve performance and/or energy efficiency. We review heterogeneous computing approaches at runtime, algorithm, programming, compiler and applicationmore » level. Further, we review both discrete and fused CPU-GPU systems; and discuss benchmark suites designed for evaluating heterogeneous computing systems (HCSs). Furthermore, we believe that this paper will provide insights into working and scope of applications of HCTs to researchers and motivate them to further harness the computational powers of CPUs and GPUs to achieve the goal of exascale performance.« less

A survey of CPU-GPU heterogeneous computing techniques

DOE Office of Scientific and Technical Information (OSTI.GOV)

Mittal, Sparsh; Vetter, Jeffrey S.

As both CPU and GPU become employed in a wide range of applications, it has been acknowledged that both of these processing units (PUs) have their unique features and strengths and hence, CPU-GPU collaboration is inevitable to achieve high-performance computing. This has motivated significant amount of research on heterogeneous computing techniques, along with the design of CPU-GPU fused chips and petascale heterogeneous supercomputers. In this paper, we survey heterogeneous computing techniques (HCTs) such as workload-partitioning which enable utilizing both CPU and GPU to improve performance and/or energy efficiency. We review heterogeneous computing approaches at runtime, algorithm, programming, compiler and applicationmore » level. Further, we review both discrete and fused CPU-GPU systems; and discuss benchmark suites designed for evaluating heterogeneous computing systems (HCSs). Furthermore, we believe that this paper will provide insights into working and scope of applications of HCTs to researchers and motivate them to further harness the computational powers of CPUs and GPUs to achieve the goal of exascale performance.« less

CPU-GPU hybrid accelerating the Zuker algorithm for RNA secondary structure prediction applications

PubMed Central

2012-01-01

Background Prediction of ribonucleic acid (RNA) secondary structure remains one of the most important research areas in bioinformatics. The Zuker algorithm is one of the most popular methods of free energy minimization for RNA secondary structure prediction. Thus far, few studies have been reported on the acceleration of the Zuker algorithm on general-purpose processors or on extra accelerators such as Field Programmable Gate-Array (FPGA) and Graphics Processing Units (GPU). To the best of our knowledge, no implementation combines both CPU and extra accelerators, such as GPUs, to accelerate the Zuker algorithm applications. Results In this paper, a CPU-GPU hybrid computing system that accelerates Zuker algorithm applications for RNA secondary structure prediction is proposed. The computing tasks are allocated between CPU and GPU for parallel cooperate execution. Performance differences between the CPU and the GPU in the task-allocation scheme are considered to obtain workload balance. To improve the hybrid system performance, the Zuker algorithm is optimally implemented with special methods for CPU and GPU architecture. Conclusions Speedup of 15.93× over optimized multi-core SIMD CPU implementation and performance advantage of 16% over optimized GPU implementation are shown in the experimental results. More than 14% of the sequences are executed on CPU in the hybrid system. The system combining CPU and GPU to accelerate the Zuker algorithm is proven to be promising and can be applied to other bioinformatics applications. PMID:22369626

MSAProbs-MPI: parallel multiple sequence aligner for distributed-memory systems.

PubMed

González-Domínguez, Jorge; Liu, Yongchao; Touriño, Juan; Schmidt, Bertil

2016-12-15

MSAProbs is a state-of-the-art protein multiple sequence alignment tool based on hidden Markov models. It can achieve high alignment accuracy at the expense of relatively long runtimes for large-scale input datasets. In this work we present MSAProbs-MPI, a distributed-memory parallel version of the multithreaded MSAProbs tool that is able to reduce runtimes by exploiting the compute capabilities of common multicore CPU clusters. Our performance evaluation on a cluster with 32 nodes (each containing two Intel Haswell processors) shows reductions in execution time of over one order of magnitude for typical input datasets. Furthermore, MSAProbs-MPI using eight nodes is faster than the GPU-accelerated QuickProbs running on a Tesla K20. Another strong point is that MSAProbs-MPI can deal with large datasets for which MSAProbs and QuickProbs might fail due to time and memory constraints, respectively. Source code in C ++ and MPI running on Linux systems as well as a reference manual are available at http://msaprobs.sourceforge.net CONTACT: jgonzalezd@udc.esSupplementary information: Supplementary data are available at Bioinformatics online. © The Author 2016. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com.

Dense GPU-enhanced surface reconstruction from stereo endoscopic images for intraoperative registration.

PubMed

Rohl, Sebastian; Bodenstedt, Sebastian; Suwelack, Stefan; Dillmann, Rudiger; Speidel, Stefanie; Kenngott, Hannes; Muller-Stich, Beat P

2012-03-01

In laparoscopic surgery, soft tissue deformations substantially change the surgical site, thus impeding the use of preoperative planning during intraoperative navigation. Extracting depth information from endoscopic images and building a surface model of the surgical field-of-view is one way to represent this constantly deforming environment. The information can then be used for intraoperative registration. Stereo reconstruction is a typical problem within computer vision. However, most of the available methods do not fulfill the specific requirements in a minimally invasive setting such as the need of real-time performance, the problem of view-dependent specular reflections and large curved areas with partly homogeneous or periodic textures and occlusions. In this paper, the authors present an approach toward intraoperative surface reconstruction based on stereo endoscopic images. The authors describe our answer to this problem through correspondence analysis, disparity correction and refinement, 3D reconstruction, point cloud smoothing and meshing. Real-time performance is achieved by implementing the algorithms on the gpu. The authors also present a new hybrid cpu-gpu algorithm that unifies the advantages of the cpu and the gpu version. In a comprehensive evaluation using in vivo data, in silico data from the literature and virtual data from a newly developed simulation environment, the cpu, the gpu, and the hybrid cpu-gpu versions of the surface reconstruction are compared to a cpu and a gpu algorithm from the literature. The recommended approach toward intraoperative surface reconstruction can be conducted in real-time depending on the image resolution (20 fps for the gpu and 14fps for the hybrid cpu-gpu version on resolution of 640 × 480). It is robust to homogeneous regions without texture, large image changes, noise or errors from camera calibration, and it reconstructs the surface down to sub millimeter accuracy. In all the experiments within the simulation environment, the mean distance to ground truth data is between 0.05 and 0.6 mm for the hybrid cpu-gpu version. The hybrid cpu-gpu algorithm shows a much more superior performance than its cpu and gpu counterpart (mean distance reduction 26% and 45%, respectively, for the experiments in the simulation environment). The recommended approach for surface reconstruction is fast, robust, and accurate. It can represent changes in the intraoperative environment and can be used to adapt a preoperative model within the surgical site by registration of these two models.

System matrix computation vs storage on GPU: A comparative study in cone beam CT.

PubMed

Matenine, Dmitri; Côté, Geoffroi; Mascolo-Fortin, Julia; Goussard, Yves; Després, Philippe

2018-02-01

Iterative reconstruction algorithms in computed tomography (CT) require a fast method for computing the intersection distances between the trajectories of photons and the object, also called ray tracing or system matrix computation. This work focused on the thin-ray model is aimed at comparing different system matrix handling strategies using graphical processing units (GPUs). In this work, the system matrix is modeled by thin rays intersecting a regular grid of box-shaped voxels, known to be an accurate representation of the forward projection operator in CT. However, an uncompressed system matrix exceeds the random access memory (RAM) capacities of typical computers by one order of magnitude or more. Considering the RAM limitations of GPU hardware, several system matrix handling methods were compared: full storage of a compressed system matrix, on-the-fly computation of its coefficients, and partial storage of the system matrix with partial on-the-fly computation. These methods were tested on geometries mimicking a cone beam CT (CBCT) acquisition of a human head. Execution times of three routines of interest were compared: forward projection, backprojection, and ordered-subsets convex (OSC) iteration. A fully stored system matrix yielded the shortest backprojection and OSC iteration times, with a 1.52× acceleration for OSC when compared to the on-the-fly approach. Nevertheless, the maximum problem size was bound by the available GPU RAM and geometrical symmetries. On-the-fly coefficient computation did not require symmetries and was shown to be the fastest for forward projection. It also offered reasonable execution times of about 176.4 ms per view per OSC iteration for a detector of 512 × 448 pixels and a volume of 384 3 voxels, using commodity GPU hardware. Partial system matrix storage has shown a performance similar to the on-the-fly approach, while still relying on symmetries. Partial system matrix storage was shown to yield the lowest relative performance. On-the-fly ray tracing was shown to be the most flexible method, yielding reasonable execution times. A fully stored system matrix allowed for the lowest backprojection and OSC iteration times and may be of interest for certain performance-oriented applications. © 2017 American Association of Physicists in Medicine.

Accelerating numerical solution of stochastic differential equations with CUDA

NASA Astrophysics Data System (ADS)

Januszewski, M.; Kostur, M.

2010-01-01

Numerical integration of stochastic differential equations is commonly used in many branches of science. In this paper we present how to accelerate this kind of numerical calculations with popular NVIDIA Graphics Processing Units using the CUDA programming environment. We address general aspects of numerical programming on stream processors and illustrate them by two examples: the noisy phase dynamics in a Josephson junction and the noisy Kuramoto model. In presented cases the measured speedup can be as high as 675× compared to a typical CPU, which corresponds to several billion integration steps per second. This means that calculations which took weeks can now be completed in less than one hour. This brings stochastic simulation to a completely new level, opening for research a whole new range of problems which can now be solved interactively. Program summaryProgram title: SDE Catalogue identifier: AEFG_v1_0 Program summary URL:http://cpc.cs.qub.ac.uk/summaries/AEFG_v1_0.html Program obtainable from: CPC Program Library, Queen's University, Belfast, N. Ireland Licensing provisions: Gnu GPL v3 No. of lines in distributed program, including test data, etc.: 978 No. of bytes in distributed program, including test data, etc.: 5905 Distribution format: tar.gz Programming language: CUDA C Computer: any system with a CUDA-compatible GPU Operating system: Linux RAM: 64 MB of GPU memory Classification: 4.3 External routines: The program requires the NVIDIA CUDA Toolkit Version 2.0 or newer and the GNU Scientific Library v1.0 or newer. Optionally gnuplot is recommended for quick visualization of the results. Nature of problem: Direct numerical integration of stochastic differential equations is a computationally intensive problem, due to the necessity of calculating multiple independent realizations of the system. We exploit the inherent parallelism of this problem and perform the calculations on GPUs using the CUDA programming environment. The GPU's ability to execute hundreds of threads simultaneously makes it possible to speed up the computation by over two orders of magnitude, compared to a typical modern CPU. Solution method: The stochastic Runge-Kutta method of the second order is applied to integrate the equation of motion. Ensemble-averaged quantities of interest are obtained through averaging over multiple independent realizations of the system. Unusual features: The numerical solution of the stochastic differential equations in question is performed on a GPU using the CUDA environment. Running time: < 1 minute

Evaluating the relation between memory and intelligence in children with learning disabilities.

PubMed

Hoerig, Dianne C; David, Andrew S; D'Amato, Rik Carl

2002-12-01

Although both intelligence tests and memory tests are commonly used in neuropsychological examinations, the relationship between memory and intelligence has not been fully explored, particularly for children having learning disabilities. Memory, or the ability to retain information, was evaluated using the Test of Memory and Learning, a recently released test that gives a comprehensive measure of global memory functioning. This, and the Wechsler Intelligence Scale for Children-Third Edition, used to assess intelligence, were given to 80 students with learning disabilities. The correlation between a global measure of memory and a global measure f intelligence was significant (r = .59), indicating that memory should be viewed as an important component when evaluating children with learning disabilities.

GPU real-time processing in NA62 trigger system

NASA Astrophysics Data System (ADS)

Ammendola, R.; Biagioni, A.; Chiozzi, S.; Cretaro, P.; Di Lorenzo, S.; Fantechi, R.; Fiorini, M.; Frezza, O.; Lamanna, G.; Lo Cicero, F.; Lonardo, A.; Martinelli, M.; Neri, I.; Paolucci, P. S.; Pastorelli, E.; Piandani, R.; Piccini, M.; Pontisso, L.; Rossetti, D.; Simula, F.; Sozzi, M.; Vicini, P.

2017-01-01

A commercial Graphics Processing Unit (GPU) is used to build a fast Level 0 (L0) trigger system tested parasitically with the TDAQ (Trigger and Data Acquisition systems) of the NA62 experiment at CERN. In particular, the parallel computing power of the GPU is exploited to perform real-time fitting in the Ring Imaging CHerenkov (RICH) detector. Direct GPU communication using a FPGA-based board has been used to reduce the data transmission latency. The performance of the system for multi-ring reconstrunction obtained during the NA62 physics run will be presented.

gWEGA: GPU-accelerated WEGA for molecular superposition and shape comparison.

PubMed

Yan, Xin; Li, Jiabo; Gu, Qiong; Xu, Jun

2014-06-05

Virtual screening of a large chemical library for drug lead identification requires searching/superimposing a large number of three-dimensional (3D) chemical structures. This article reports a graphic processing unit (GPU)-accelerated weighted Gaussian algorithm (gWEGA) that expedites shape or shape-feature similarity score-based virtual screening. With 86 GPU nodes (each node has one GPU card), gWEGA can screen 110 million conformations derived from an entire ZINC drug-like database with diverse antidiabetic agents as query structures within 2 s (i.e., screening more than 55 million conformations per second). The rapid screening speed was accomplished through the massive parallelization on multiple GPU nodes and rapid prescreening of 3D structures (based on their shape descriptors and pharmacophore feature compositions). Copyright © 2014 Wiley Periodicals, Inc.

Graphics Processing Unit Acceleration of Gyrokinetic Turbulence Simulations

NASA Astrophysics Data System (ADS)

Hause, Benjamin; Parker, Scott

2012-10-01

We find a substantial increase in on-node performance using Graphics Processing Unit (GPU) acceleration in gyrokinetic delta-f particle-in-cell simulation. Optimization is performed on a two-dimensional slab gyrokinetic particle simulation using the Portland Group Fortran compiler with the GPU accelerator compiler directives. We have implemented the GPU acceleration on a Core I7 gaming PC with a NVIDIA GTX 580 GPU. We find comparable, or better, acceleration relative to the NERSC DIRAC cluster with the NVIDIA Tesla C2050 computing processor. The Tesla C 2050 is about 2.6 times more expensive than the GTX 580 gaming GPU. Optimization strategies and comparisons between DIRAC and the gaming PC will be presented. We will also discuss progress on optimizing the comprehensive three dimensional general geometry GEM code.

Understanding GPU Power. A Survey of Profiling, Modeling, and Simulation Methods

DOE PAGES

Bridges, Robert A.; Imam, Neena; Mintz, Tiffany M.

2016-09-01

Modern graphics processing units (GPUs) have complex architectures that admit exceptional performance and energy efficiency for high throughput applications.Though GPUs consume large amounts of power, their use for high throughput applications facilitate state-of-the-art energy efficiency and performance. Consequently, continued development relies on understanding their power consumption. Our work is a survey of GPU power modeling and profiling methods with increased detail on noteworthy efforts. Moreover, as direct measurement of GPU power is necessary for model evaluation and parameter initiation, internal and external power sensors are discussed. Hardware counters, which are low-level tallies of hardware events, share strong correlation to powermore » use and performance. Statistical correlation between power and performance counters has yielded worthwhile GPU power models, yet the complexity inherent to GPU architectures presents new hurdles for power modeling. Developments and challenges of counter-based GPU power modeling is discussed. Often building on the counter-based models, research efforts for GPU power simulation, which make power predictions from input code and hardware knowledge, provide opportunities for optimization in programming or architectural design. Noteworthy strides in power simulations for GPUs are included along with their performance or functional simulator counterparts when appropriate. Lastly, possible directions for future research are discussed.« less

GPU-based prompt gamma ray imaging from boron neutron capture therapy.

PubMed

Yoon, Do-Kun; Jung, Joo-Young; Jo Hong, Key; Sil Lee, Keum; Suk Suh, Tae

2015-01-01

The purpose of this research is to perform the fast reconstruction of a prompt gamma ray image using a graphics processing unit (GPU) computation from boron neutron capture therapy (BNCT) simulations. To evaluate the accuracy of the reconstructed image, a phantom including four boron uptake regions (BURs) was used in the simulation. After the Monte Carlo simulation of the BNCT, the modified ordered subset expectation maximization reconstruction algorithm using the GPU computation was used to reconstruct the images with fewer projections. The computation times for image reconstruction were compared between the GPU and the central processing unit (CPU). Also, the accuracy of the reconstructed image was evaluated by a receiver operating characteristic (ROC) curve analysis. The image reconstruction time using the GPU was 196 times faster than the conventional reconstruction time using the CPU. For the four BURs, the area under curve values from the ROC curve were 0.6726 (A-region), 0.6890 (B-region), 0.7384 (C-region), and 0.8009 (D-region). The tomographic image using the prompt gamma ray event from the BNCT simulation was acquired using the GPU computation in order to perform a fast reconstruction during treatment. The authors verified the feasibility of the prompt gamma ray image reconstruction using the GPU computation for BNCT simulations.

Understanding GPU Power. A Survey of Profiling, Modeling, and Simulation Methods

DOE Office of Scientific and Technical Information (OSTI.GOV)

Bridges, Robert A.; Imam, Neena; Mintz, Tiffany M.

Modern graphics processing units (GPUs) have complex architectures that admit exceptional performance and energy efficiency for high throughput applications.Though GPUs consume large amounts of power, their use for high throughput applications facilitate state-of-the-art energy efficiency and performance. Consequently, continued development relies on understanding their power consumption. Our work is a survey of GPU power modeling and profiling methods with increased detail on noteworthy efforts. Moreover, as direct measurement of GPU power is necessary for model evaluation and parameter initiation, internal and external power sensors are discussed. Hardware counters, which are low-level tallies of hardware events, share strong correlation to powermore » use and performance. Statistical correlation between power and performance counters has yielded worthwhile GPU power models, yet the complexity inherent to GPU architectures presents new hurdles for power modeling. Developments and challenges of counter-based GPU power modeling is discussed. Often building on the counter-based models, research efforts for GPU power simulation, which make power predictions from input code and hardware knowledge, provide opportunities for optimization in programming or architectural design. Noteworthy strides in power simulations for GPUs are included along with their performance or functional simulator counterparts when appropriate. Lastly, possible directions for future research are discussed.« less

Parallelized computation for computer simulation of electrocardiograms using personal computers with multi-core CPU and general-purpose GPU.

PubMed

Shen, Wenfeng; Wei, Daming; Xu, Weimin; Zhu, Xin; Yuan, Shizhong

2010-10-01

Biological computations like electrocardiological modelling and simulation usually require high-performance computing environments. This paper introduces an implementation of parallel computation for computer simulation of electrocardiograms (ECGs) in a personal computer environment with an Intel CPU of Core (TM) 2 Quad Q6600 and a GPU of Geforce 8800GT, with software support by OpenMP and CUDA. It was tested in three parallelization device setups: (a) a four-core CPU without a general-purpose GPU, (b) a general-purpose GPU plus 1 core of CPU, and (c) a four-core CPU plus a general-purpose GPU. To effectively take advantage of a multi-core CPU and a general-purpose GPU, an algorithm based on load-prediction dynamic scheduling was developed and applied to setting (c). In the simulation with 1600 time steps, the speedup of the parallel computation as compared to the serial computation was 3.9 in setting (a), 16.8 in setting (b), and 20.0 in setting (c). This study demonstrates that a current PC with a multi-core CPU and a general-purpose GPU provides a good environment for parallel computations in biological modelling and simulation studies. Copyright 2010 Elsevier Ireland Ltd. All rights reserved.

Research on fast Fourier transforms algorithm of huge remote sensing image technology with GPU and partitioning technology.

PubMed

Yang, Xue; Li, Xue-You; Li, Jia-Guo; Ma, Jun; Zhang, Li; Yang, Jan; Du, Quan-Ye

2014-02-01

Fast Fourier transforms (FFT) is a basic approach to remote sensing image processing. With the improvement of capacity of remote sensing image capture with the features of hyperspectrum, high spatial resolution and high temporal resolution, how to use FFT technology to efficiently process huge remote sensing image becomes the critical step and research hot spot of current image processing technology. FFT algorithm, one of the basic algorithms of image processing, can be used for stripe noise removal, image compression, image registration, etc. in processing remote sensing image. CUFFT function library is the FFT algorithm library based on CPU and FFTW. FFTW is a FFT algorithm developed based on CPU in PC platform, and is currently the fastest CPU based FFT algorithm function library. However there is a common problem that once the available memory or memory is less than the capacity of image, there will be out of memory or memory overflow when using the above two methods to realize image FFT arithmetic. To address this problem, a CPU and partitioning technology based Huge Remote Fast Fourier Transform (HRFFT) algorithm is proposed in this paper. By improving the FFT algorithm in CUFFT function library, the problem of out of memory and memory overflow is solved. Moreover, this method is proved rational by experiment combined with the CCD image of HJ-1A satellite. When applied to practical image processing, it improves effect of the image processing, speeds up the processing, which saves the time of computation and achieves sound result.

Neural Global Pattern Similarity Underlies True and False Memories.

PubMed

Ye, Zhifang; Zhu, Bi; Zhuang, Liping; Lu, Zhonglin; Chen, Chuansheng; Xue, Gui

2016-06-22

The neural processes giving rise to human memory strength signals remain poorly understood. Inspired by formal computational models that posit a central role of global matching in memory strength, we tested a novel hypothesis that the strengths of both true and false memories arise from the global similarity of an item's neural activation pattern during retrieval to that of all the studied items during encoding (i.e., the encoding-retrieval neural global pattern similarity [ER-nGPS]). We revealed multiple ER-nGPS signals that carried distinct information and contributed differentially to true and false memories: Whereas the ER-nGPS in the parietal regions reflected semantic similarity and was scaled with the recognition strengths of both true and false memories, ER-nGPS in the visual cortex contributed solely to true memory. Moreover, ER-nGPS differences between the parietal and visual cortices were correlated with frontal monitoring processes. By combining computational and neuroimaging approaches, our results advance a mechanistic understanding of memory strength in recognition. What neural processes give rise to memory strength signals, and lead to our conscious feelings of familiarity? Using fMRI, we found that the memory strength of a given item depends not only on how it was encoded during learning, but also on the similarity of its neural representation with other studied items. The global neural matching signal, mainly in the parietal lobule, could account for the memory strengths of both studied and unstudied items. Interestingly, a different global matching signal, originated from the visual cortex, could distinguish true from false memories. The findings reveal multiple neural mechanisms underlying the memory strengths of events registered in the brain. Copyright © 2016 the authors 0270-6474/16/366792-11$15.00/0.

Collaborating CPU and GPU for large-scale high-order CFD simulations with complex grids on the TianHe-1A supercomputer

NASA Astrophysics Data System (ADS)

Xu, Chuanfu; Deng, Xiaogang; Zhang, Lilun; Fang, Jianbin; Wang, Guangxue; Jiang, Yi; Cao, Wei; Che, Yonggang; Wang, Yongxian; Wang, Zhenghua; Liu, Wei; Cheng, Xinghua

2014-12-01

Programming and optimizing complex, real-world CFD codes on current many-core accelerated HPC systems is very challenging, especially when collaborating CPUs and accelerators to fully tap the potential of heterogeneous systems. In this paper, with a tri-level hybrid and heterogeneous programming model using MPI + OpenMP + CUDA, we port and optimize our high-order multi-block structured CFD software HOSTA on the GPU-accelerated TianHe-1A supercomputer. HOSTA adopts two self-developed high-order compact definite difference schemes WCNS and HDCS that can simulate flows with complex geometries. We present a dual-level parallelization scheme for efficient multi-block computation on GPUs and perform particular kernel optimizations for high-order CFD schemes. The GPU-only approach achieves a speedup of about 1.3 when comparing one Tesla M2050 GPU with two Xeon X5670 CPUs. To achieve a greater speedup, we collaborate CPU and GPU for HOSTA instead of using a naive GPU-only approach. We present a novel scheme to balance the loads between the store-poor GPU and the store-rich CPU. Taking CPU and GPU load balance into account, we improve the maximum simulation problem size per TianHe-1A node for HOSTA by 2.3×, meanwhile the collaborative approach can improve the performance by around 45% compared to the GPU-only approach. Further, to scale HOSTA on TianHe-1A, we propose a gather/scatter optimization to minimize PCI-e data transfer times for ghost and singularity data of 3D grid blocks, and overlap the collaborative computation and communication as far as possible using some advanced CUDA and MPI features. Scalability tests show that HOSTA can achieve a parallel efficiency of above 60% on 1024 TianHe-1A nodes. With our method, we have successfully simulated an EET high-lift airfoil configuration containing 800M cells and China's large civil airplane configuration containing 150M cells. To our best knowledge, those are the largest-scale CPU-GPU collaborative simulations that solve realistic CFD problems with both complex configurations and high-order schemes.

Collaborating CPU and GPU for large-scale high-order CFD simulations with complex grids on the TianHe-1A supercomputer

DOE Office of Scientific and Technical Information (OSTI.GOV)

Xu, Chuanfu, E-mail: xuchuanfu@nudt.edu.cn; Deng, Xiaogang; Zhang, Lilun

Programming and optimizing complex, real-world CFD codes on current many-core accelerated HPC systems is very challenging, especially when collaborating CPUs and accelerators to fully tap the potential of heterogeneous systems. In this paper, with a tri-level hybrid and heterogeneous programming model using MPI + OpenMP + CUDA, we port and optimize our high-order multi-block structured CFD software HOSTA on the GPU-accelerated TianHe-1A supercomputer. HOSTA adopts two self-developed high-order compact definite difference schemes WCNS and HDCS that can simulate flows with complex geometries. We present a dual-level parallelization scheme for efficient multi-block computation on GPUs and perform particular kernel optimizations formore » high-order CFD schemes. The GPU-only approach achieves a speedup of about 1.3 when comparing one Tesla M2050 GPU with two Xeon X5670 CPUs. To achieve a greater speedup, we collaborate CPU and GPU for HOSTA instead of using a naive GPU-only approach. We present a novel scheme to balance the loads between the store-poor GPU and the store-rich CPU. Taking CPU and GPU load balance into account, we improve the maximum simulation problem size per TianHe-1A node for HOSTA by 2.3×, meanwhile the collaborative approach can improve the performance by around 45% compared to the GPU-only approach. Further, to scale HOSTA on TianHe-1A, we propose a gather/scatter optimization to minimize PCI-e data transfer times for ghost and singularity data of 3D grid blocks, and overlap the collaborative computation and communication as far as possible using some advanced CUDA and MPI features. Scalability tests show that HOSTA can achieve a parallel efficiency of above 60% on 1024 TianHe-1A nodes. With our method, we have successfully simulated an EET high-lift airfoil configuration containing 800M cells and China's large civil airplane configuration containing 150M cells. To our best knowledge, those are the largest-scale CPU–GPU collaborative simulations that solve realistic CFD problems with both complex configurations and high-order schemes.« less

cellGPU: Massively parallel simulations of dynamic vertex models

NASA Astrophysics Data System (ADS)

Sussman, Daniel M.

2017-10-01

Vertex models represent confluent tissue by polygonal or polyhedral tilings of space, with the individual cells interacting via force laws that depend on both the geometry of the cells and the topology of the tessellation. This dependence on the connectivity of the cellular network introduces several complications to performing molecular-dynamics-like simulations of vertex models, and in particular makes parallelizing the simulations difficult. cellGPU addresses this difficulty and lays the foundation for massively parallelized, GPU-based simulations of these models. This article discusses its implementation for a pair of two-dimensional models, and compares the typical performance that can be expected between running cellGPU entirely on the CPU versus its performance when running on a range of commercial and server-grade graphics cards. By implementing the calculation of topological changes and forces on cells in a highly parallelizable fashion, cellGPU enables researchers to simulate time- and length-scales previously inaccessible via existing single-threaded CPU implementations. Program Files doi:http://dx.doi.org/10.17632/6j2cj29t3r.1 Licensing provisions: MIT Programming language: CUDA/C++ Nature of problem: Simulations of off-lattice "vertex models" of cells, in which the interaction forces depend on both the geometry and the topology of the cellular aggregate. Solution method: Highly parallelized GPU-accelerated dynamical simulations in which the force calculations and the topological features can be handled on either the CPU or GPU. Additional comments: The code is hosted at https://gitlab.com/dmsussman/cellGPU, with documentation additionally maintained at http://dmsussman.gitlab.io/cellGPUdocumentation

Testing and Validating Gadget2 for GPUs

NASA Astrophysics Data System (ADS)

Wibking, Benjamin; Holley-Bockelmann, K.; Berlind, A. A.

2013-01-01

We are currently upgrading a version of Gadget2 (Springel et al., 2005) that is optimized for NVIDIA's CUDA GPU architecture (Frigaard, unpublished) to work with the latest libraries and graphics cards. Preliminary tests of its performance indicate a ~40x speedup in the particle force tree approximation calculation, with overall speedup of 5-10x for cosmological simulations run with GPUs compared to running on the same CPU cores without GPU acceleration. We believe this speedup can be reasonably increased by an additional factor of two with futher optimization, including overlap of computation on CPU and GPU. Tests of single-precision GPU numerical fidelity currently indicate accuracy of the mass function and the spectral power density to within a few percent of extended-precision CPU results with the unmodified form of Gadget. Additionally, we plan to test and optimize the GPU code for Millenium-scale "grand challenge" simulations of >10^9 particles, a scale that has been previously untested with this code, with the aid of the NSF XSEDE flagship GPU-based supercomputing cluster codenamed "Keeneland." Current work involves additional validation of numerical results, extending the numerical precision of the GPU calculations to double precision, and evaluating performance/accuracy tradeoffs. We believe that this project, if successful, will yield substantial computational performance benefits to the N-body research community as the next generation of GPU supercomputing resources becomes available, both increasing the electrical power efficiency of ever-larger computations (making simulations possible a decade from now at scales and resolutions unavailable today) and accelerating the pace of research in the field.

GPU-accelerated automatic identification of robust beam setups for proton and carbon-ion radiotherapy

NASA Astrophysics Data System (ADS)

Ammazzalorso, F.; Bednarz, T.; Jelen, U.

2014-03-01

We demonstrate acceleration on graphic processing units (GPU) of automatic identification of robust particle therapy beam setups, minimizing negative dosimetric effects of Bragg peak displacement caused by treatment-time patient positioning errors. Our particle therapy research toolkit, RobuR, was extended with OpenCL support and used to implement calculation on GPU of the Port Homogeneity Index, a metric scoring irradiation port robustness through analysis of tissue density patterns prior to dose optimization and computation. Results were benchmarked against an independent native CPU implementation. Numerical results were in agreement between the GPU implementation and native CPU implementation. For 10 skull base cases, the GPU-accelerated implementation was employed to select beam setups for proton and carbon ion treatment plans, which proved to be dosimetrically robust, when recomputed in presence of various simulated positioning errors. From the point of view of performance, average running time on the GPU decreased by at least one order of magnitude compared to the CPU, rendering the GPU-accelerated analysis a feasible step in a clinical treatment planning interactive session. In conclusion, selection of robust particle therapy beam setups can be effectively accelerated on a GPU and become an unintrusive part of the particle therapy treatment planning workflow. Additionally, the speed gain opens new usage scenarios, like interactive analysis manipulation (e.g. constraining of some setup) and re-execution. Finally, through OpenCL portable parallelism, the new implementation is suitable also for CPU-only use, taking advantage of multiple cores, and can potentially exploit types of accelerators other than GPUs.

Performance evaluation of throughput computing workloads using multi-core processors and graphics processors

NASA Astrophysics Data System (ADS)

Dave, Gaurav P.; Sureshkumar, N.; Blessy Trencia Lincy, S. S.

2017-11-01

Current trend in processor manufacturing focuses on multi-core architectures rather than increasing the clock speed for performance improvement. Graphic processors have become as commodity hardware for providing fast co-processing in computer systems. Developments in IoT, social networking web applications, big data created huge demand for data processing activities and such kind of throughput intensive applications inherently contains data level parallelism which is more suited for SIMD architecture based GPU. This paper reviews the architectural aspects of multi/many core processors and graphics processors. Different case studies are taken to compare performance of throughput computing applications using shared memory programming in OpenMP and CUDA API based programming.

GPU-Based Point Cloud Superpositioning for Structural Comparisons of Protein Binding Sites.

PubMed

Leinweber, Matthias; Fober, Thomas; Freisleben, Bernd

2018-01-01

In this paper, we present a novel approach to solve the labeled point cloud superpositioning problem for performing structural comparisons of protein binding sites. The solution is based on a parallel evolution strategy that operates on large populations and runs on GPU hardware. The proposed evolution strategy reduces the likelihood of getting stuck in a local optimum of the multimodal real-valued optimization problem represented by labeled point cloud superpositioning. The performance of the GPU-based parallel evolution strategy is compared to a previously proposed CPU-based sequential approach for labeled point cloud superpositioning, indicating that the GPU-based parallel evolution strategy leads to qualitatively better results and significantly shorter runtimes, with speed improvements of up to a factor of 1,500 for large populations. Binary classification tests based on the ATP, NADH, and FAD protein subsets of CavBase, a database containing putative binding sites, show average classification rate improvements from about 92 percent (CPU) to 96 percent (GPU). Further experiments indicate that the proposed GPU-based labeled point cloud superpositioning approach can be superior to traditional protein comparison approaches based on sequence alignments.

irGPU.proton.Net: Irregular strong charge interaction networks of protonatable groups in protein molecules--a GPU solver using the fast multipole method and statistical thermodynamics.

PubMed

Kantardjiev, Alexander A

2015-04-05

A cluster of strongly interacting ionization groups in protein molecules with irregular ionization behavior is suggestive for specific structure-function relationship. However, their computational treatment is unconventional (e.g., lack of convergence in naive self-consistent iterative algorithm). The stringent evaluation requires evaluation of Boltzmann averaged statistical mechanics sums and electrostatic energy estimation for each microstate. irGPU: Irregular strong interactions in proteins--a GPU solver is novel solution to a versatile problem in protein biophysics--atypical protonation behavior of coupled groups. The computational severity of the problem is alleviated by parallelization (via GPU kernels) which is applied for the electrostatic interaction evaluation (including explicit electrostatics via the fast multipole method) as well as statistical mechanics sums (partition function) estimation. Special attention is given to the ease of the service and encapsulation of theoretical details without sacrificing rigor of computational procedures. irGPU is not just a solution-in-principle but a promising practical application with potential to entice community into deeper understanding of principles governing biomolecule mechanisms. © 2015 Wiley Periodicals, Inc.

Optimizing Tensor Contraction Expressions for Hybrid CPU-GPU Execution

DOE Office of Scientific and Technical Information (OSTI.GOV)

Ma, Wenjing; Krishnamoorthy, Sriram; Villa, Oreste

2013-03-01

Tensor contractions are generalized multidimensional matrix multiplication operations that widely occur in quantum chemistry. Efficient execution of tensor contractions on Graphics Processing Units (GPUs) requires several challenges to be addressed, including index permutation and small dimension-sizes reducing thread block utilization. Moreover, to apply the same optimizations to various expressions, we need a code generation tool. In this paper, we present our approach to automatically generate CUDA code to execute tensor contractions on GPUs, including management of data movement between CPU and GPU. To evaluate our tool, GPU-enabled code is generated for the most expensive contractions in CCSD(T), a key coupledmore » cluster method, and incorporated into NWChem, a popular computational chemistry suite. For this method, we demonstrate speedup over a factor of 8.4 using one GPU (instead of one core per node) and over 2.6 when utilizing the entire system using hybrid CPU+GPU solution with 2 GPUs and 5 cores (instead of 7 cores per node). Finally, we analyze the implementation behavior on future GPU systems.« less

A Hybrid CPU/GPU Pattern-Matching Algorithm for Deep Packet Inspection

PubMed Central

Chen, Yaw-Chung

2015-01-01

The large quantities of data now being transferred via high-speed networks have made deep packet inspection indispensable for security purposes. Scalable and low-cost signature-based network intrusion detection systems have been developed for deep packet inspection for various software platforms. Traditional approaches that only involve central processing units (CPUs) are now considered inadequate in terms of inspection speed. Graphic processing units (GPUs) have superior parallel processing power, but transmission bottlenecks can reduce optimal GPU efficiency. In this paper we describe our proposal for a hybrid CPU/GPU pattern-matching algorithm (HPMA) that divides and distributes the packet-inspecting workload between a CPU and GPU. All packets are initially inspected by the CPU and filtered using a simple pre-filtering algorithm, and packets that might contain malicious content are sent to the GPU for further inspection. Test results indicate that in terms of random payload traffic, the matching speed of our proposed algorithm was 3.4 times and 2.7 times faster than those of the AC-CPU and AC-GPU algorithms, respectively. Further, HPMA achieved higher energy efficiency than the other tested algorithms. PMID:26437335

A Hybrid CPU/GPU Pattern-Matching Algorithm for Deep Packet Inspection.

PubMed

Lee, Chun-Liang; Lin, Yi-Shan; Chen, Yaw-Chung

2015-01-01

The large quantities of data now being transferred via high-speed networks have made deep packet inspection indispensable for security purposes. Scalable and low-cost signature-based network intrusion detection systems have been developed for deep packet inspection for various software platforms. Traditional approaches that only involve central processing units (CPUs) are now considered inadequate in terms of inspection speed. Graphic processing units (GPUs) have superior parallel processing power, but transmission bottlenecks can reduce optimal GPU efficiency. In this paper we describe our proposal for a hybrid CPU/GPU pattern-matching algorithm (HPMA) that divides and distributes the packet-inspecting workload between a CPU and GPU. All packets are initially inspected by the CPU and filtered using a simple pre-filtering algorithm, and packets that might contain malicious content are sent to the GPU for further inspection. Test results indicate that in terms of random payload traffic, the matching speed of our proposed algorithm was 3.4 times and 2.7 times faster than those of the AC-CPU and AC-GPU algorithms, respectively. Further, HPMA achieved higher energy efficiency than the other tested algorithms.

Fast 3D elastic micro-seismic source location using new GPU features

NASA Astrophysics Data System (ADS)

Xue, Qingfeng; Wang, Yibo; Chang, Xu

2016-12-01

In this paper, we describe new GPU features and their applications in passive seismic - micro-seismic location. Locating micro-seismic events is quite important in seismic exploration, especially when searching for unconventional oil and gas resources. Different from the traditional ray-based methods, the wave equation method, such as the method we use in our paper, has a remarkable advantage in adapting to low signal-to-noise ratio conditions and does not need a person to select the data. However, because it has a conspicuous deficiency due to its computation cost, these methods are not widely used in industrial fields. To make the method useful, we implement imaging-like wave equation micro-seismic location in a 3D elastic media and use GPU to accelerate our algorithm. We also introduce some new GPU features into the implementation to solve the data transfer and GPU utilization problems. Numerical and field data experiments show that our method can achieve a more than 30% performance improvement in GPU implementation just by using these new features.

GPU-accelerated Tersoff potentials for massively parallel Molecular Dynamics simulations

NASA Astrophysics Data System (ADS)

Nguyen, Trung Dac

2017-03-01

The Tersoff potential is one of the empirical many-body potentials that has been widely used in simulation studies at atomic scales. Unlike pair-wise potentials, the Tersoff potential involves three-body terms, which require much more arithmetic operations and data dependency. In this contribution, we have implemented the GPU-accelerated version of several variants of the Tersoff potential for LAMMPS, an open-source massively parallel Molecular Dynamics code. Compared to the existing MPI implementation in LAMMPS, the GPU implementation exhibits a better scalability and offers a speedup of 2.2X when run on 1000 compute nodes on the Titan supercomputer. On a single node, the speedup ranges from 2.0 to 8.0 times, depending on the number of atoms per GPU and hardware configurations. The most notable features of our GPU-accelerated version include its design for MPI/accelerator heterogeneous parallelism, its compatibility with other functionalities in LAMMPS, its ability to give deterministic results and to support both NVIDIA CUDA- and OpenCL-enabled accelerators. Our implementation is now part of the GPU package in LAMMPS and accessible for public use.

Implementation of Multipattern String Matching Accelerated with GPU for Intrusion Detection System

NASA Astrophysics Data System (ADS)

Nehemia, Rangga; Lim, Charles; Galinium, Maulahikmah; Rinaldi Widianto, Ahmad

2017-04-01

As Internet-related security threats continue to increase in terms of volume and sophistication, existing Intrusion Detection System is also being challenged to cope with the current Internet development. Multi Pattern String Matching algorithm accelerated with Graphical Processing Unit is being utilized to improve the packet scanning performance of the IDS. This paper implements a Multi Pattern String Matching algorithm, also called Parallel Failureless Aho Corasick accelerated with GPU to improve the performance of IDS. OpenCL library is used to allow the IDS to support various GPU, including popular GPU such as NVIDIA and AMD, used in our research. The experiment result shows that the application of Multi Pattern String Matching using GPU accelerated platform provides a speed up, by up to 141% in term of throughput compared to the previous research.

Multi-GPU accelerated three-dimensional FDTD method for electromagnetic simulation.

PubMed

Nagaoka, Tomoaki; Watanabe, Soichi

2011-01-01

Numerical simulation with a numerical human model using the finite-difference time domain (FDTD) method has recently been performed in a number of fields in biomedical engineering. To improve the method's calculation speed and realize large-scale computing with the numerical human model, we adapt three-dimensional FDTD code to a multi-GPU environment using Compute Unified Device Architecture (CUDA). In this study, we used NVIDIA Tesla C2070 as GPGPU boards. The performance of multi-GPU is evaluated in comparison with that of a single GPU and vector supercomputer. The calculation speed with four GPUs was approximately 3.5 times faster than with a single GPU, and was slightly (approx. 1.3 times) slower than with the supercomputer. Calculation speed of the three-dimensional FDTD method using GPUs can significantly improve with an expanding number of GPUs.

Efficient implementation of the 3D-DDA ray traversal algorithm on GPU and its application in radiation dose calculation.

PubMed

Xiao, Kai; Chen, Danny Z; Hu, X Sharon; Zhou, Bo

2012-12-01

The three-dimensional digital differential analyzer (3D-DDA) algorithm is a widely used ray traversal method, which is also at the core of many convolution∕superposition (C∕S) dose calculation approaches. However, porting existing C∕S dose calculation methods onto graphics processing unit (GPU) has brought challenges to retaining the efficiency of this algorithm. In particular, straightforward implementation of the original 3D-DDA algorithm inflicts a lot of branch divergence which conflicts with the GPU programming model and leads to suboptimal performance. In this paper, an efficient GPU implementation of the 3D-DDA algorithm is proposed, which effectively reduces such branch divergence and improves performance of the C∕S dose calculation programs running on GPU. The main idea of the proposed method is to convert a number of conditional statements in the original 3D-DDA algorithm into a set of simple operations (e.g., arithmetic, comparison, and logic) which are better supported by the GPU architecture. To verify and demonstrate the performance improvement, this ray traversal method was integrated into a GPU-based collapsed cone convolution∕superposition (CCCS) dose calculation program. The proposed method has been tested using a water phantom and various clinical cases on an NVIDIA GTX570 GPU. The CCCS dose calculation program based on the efficient 3D-DDA ray traversal implementation runs 1.42 ∼ 2.67× faster than the one based on the original 3D-DDA implementation, without losing any accuracy. The results show that the proposed method can effectively reduce branch divergence in the original 3D-DDA ray traversal algorithm and improve the performance of the CCCS program running on GPU. Considering the wide utilization of the 3D-DDA algorithm, various applications can benefit from this implementation method.

GPU-accelerated Monte Carlo convolution/superposition implementation for dose calculation.

PubMed

Zhou, Bo; Yu, Cedric X; Chen, Danny Z; Hu, X Sharon

2010-11-01

Dose calculation is a key component in radiation treatment planning systems. Its performance and accuracy are crucial to the quality of treatment plans as emerging advanced radiation therapy technologies are exerting ever tighter constraints on dose calculation. A common practice is to choose either a deterministic method such as the convolution/superposition (CS) method for speed or a Monte Carlo (MC) method for accuracy. The goal of this work is to boost the performance of a hybrid Monte Carlo convolution/superposition (MCCS) method by devising a graphics processing unit (GPU) implementation so as to make the method practical for day-to-day usage. Although the MCCS algorithm combines the merits of MC fluence generation and CS fluence transport, it is still not fast enough to be used as a day-to-day planning tool. To alleviate the speed issue of MC algorithms, the authors adopted MCCS as their target method and implemented a GPU-based version. In order to fully utilize the GPU computing power, the MCCS algorithm is modified to match the GPU hardware architecture. The performance of the authors' GPU-based implementation on an Nvidia GTX260 card is compared to a multithreaded software implementation on a quad-core system. A speedup in the range of 6.7-11.4x is observed for the clinical cases used. The less than 2% statistical fluctuation also indicates that the accuracy of the authors' GPU-based implementation is in good agreement with the results from the quad-core CPU implementation. This work shows that GPU is a feasible and cost-efficient solution compared to other alternatives such as using cluster machines or field-programmable gate arrays for satisfying the increasing demands on computation speed and accuracy of dose calculation. But there are also inherent limitations of using GPU for accelerating MC-type applications, which are also analyzed in detail in this article.

SU-E-T-423: Fast Photon Convolution Calculation with a 3D-Ideal Kernel On the GPU

DOE Office of Scientific and Technical Information (OSTI.GOV)

Moriya, S; Sato, M; Tachibana, H

Purpose: The calculation time is a trade-off for improving the accuracy of convolution dose calculation with fine calculation spacing of the KERMA kernel. We investigated to accelerate the convolution calculation using an ideal kernel on the Graphic Processing Units (GPU). Methods: The calculation was performed on the AMD graphics hardware of Dual FirePro D700 and our algorithm was implemented using the Aparapi that convert Java bytecode to OpenCL. The process of dose calculation was separated with the TERMA and KERMA steps. The dose deposited at the coordinate (x, y, z) was determined in the process. In the dose calculation runningmore » on the central processing unit (CPU) of Intel Xeon E5, the calculation loops were performed for all calculation points. On the GPU computation, all of the calculation processes for the points were sent to the GPU and the multi-thread computation was done. In this study, the dose calculation was performed in a water equivalent homogeneous phantom with 150{sup 3} voxels (2 mm calculation grid) and the calculation speed on the GPU to that on the CPU and the accuracy of PDD were compared. Results: The calculation time for the GPU and the CPU were 3.3 sec and 4.4 hour, respectively. The calculation speed for the GPU was 4800 times faster than that for the CPU. The PDD curve for the GPU was perfectly matched to that for the CPU. Conclusion: The convolution calculation with the ideal kernel on the GPU was clinically acceptable for time and may be more accurate in an inhomogeneous region. Intensity modulated arc therapy needs dose calculations for different gantry angles at many control points. Thus, it would be more practical that the kernel uses a coarse spacing technique if the calculation is faster while keeping the similar accuracy to a current treatment planning system.« less

NLSEmagic: Nonlinear Schrödinger equation multi-dimensional Matlab-based GPU-accelerated integrators using compact high-order schemes

NASA Astrophysics Data System (ADS)

Caplan, R. M.

2013-04-01

We present a simple to use, yet powerful code package called NLSEmagic to numerically integrate the nonlinear Schrödinger equation in one, two, and three dimensions. NLSEmagic is a high-order finite-difference code package which utilizes graphic processing unit (GPU) parallel architectures. The codes running on the GPU are many times faster than their serial counterparts, and are much cheaper to run than on standard parallel clusters. The codes are developed with usability and portability in mind, and therefore are written to interface with MATLAB utilizing custom GPU-enabled C codes with the MEX-compiler interface. The packages are freely distributed, including user manuals and set-up files. Catalogue identifier: AEOJ_v1_0 Program summary URL:http://cpc.cs.qub.ac.uk/summaries/AEOJ_v1_0.html Program obtainable from: CPC Program Library, Queen’s University, Belfast, N. Ireland Licensing provisions: Standard CPC licence, http://cpc.cs.qub.ac.uk/licence/licence.html No. of lines in distributed program, including test data, etc.: 124453 No. of bytes in distributed program, including test data, etc.: 4728604 Distribution format: tar.gz Programming language: C, CUDA, MATLAB. Computer: PC, MAC. Operating system: Windows, MacOS, Linux. Has the code been vectorized or parallelized?: Yes. Number of processors used: Single CPU, number of GPU processors dependent on chosen GPU card (max is currently 3072 cores on GeForce GTX 690). Supplementary material: Setup guide, Installation guide. RAM: Highly dependent on dimensionality and grid size. For typical medium-large problem size in three dimensions, 4GB is sufficient. Keywords: Nonlinear Schröodinger Equation, GPU, high-order finite difference, Bose-Einstien condensates. Classification: 4.3, 7.7. Nature of problem: Integrate solutions of the time-dependent one-, two-, and three-dimensional cubic nonlinear Schrödinger equation. Solution method: The integrators utilize a fully-explicit fourth-order Runge-Kutta scheme in time and both second- and fourth-order differencing in space. The integrators are written to run on NVIDIA GPUs and are interfaced with MATLAB including built-in visualization and analysis tools. Restrictions: The main restriction for the GPU integrators is the amount of RAM on the GPU as the code is currently only designed for running on a single GPU. Unusual features: Ability to visualize real-time simulations through the interaction of MATLAB and the compiled GPU integrators. Additional comments: Setup guide and Installation guide provided. Program has a dedicated web site at www.nlsemagic.com. Running time: A three-dimensional run with a grid dimension of 87×87×203 for 3360 time steps (100 non-dimensional time units) takes about one and a half minutes on a GeForce GTX 580 GPU card.

Global Neural Pattern Similarity as a Common Basis for Categorization and Recognition Memory

PubMed Central

Xue, Gui; Love, Bradley C.; Preston, Alison R.; Poldrack, Russell A.

2014-01-01

Familiarity, or memory strength, is a central construct in models of cognition. In previous categorization and long-term memory research, correlations have been found between psychological measures of memory strength and activation in the medial temporal lobes (MTLs), which suggests a common neural locus for memory strength. However, activation alone is insufficient for determining whether the same mechanisms underlie neural function across domains. Guided by mathematical models of categorization and long-term memory, we develop a theory and a method to test whether memory strength arises from the global similarity among neural representations. In human subjects, we find significant correlations between global similarity among activation patterns in the MTLs and both subsequent memory confidence in a recognition memory task and model-based measures of memory strength in a category learning task. Our work bridges formal cognitive theories and neuroscientific models by illustrating that the same global similarity computations underlie processing in multiple cognitive domains. Moreover, by establishing a link between neural similarity and psychological memory strength, our findings suggest that there may be an isomorphism between psychological and neural representational spaces that can be exploited to test cognitive theories at both the neural and behavioral levels. PMID:24872552

GPU-based prompt gamma ray imaging from boron neutron capture therapy

DOE Office of Scientific and Technical Information (OSTI.GOV)

Yoon, Do-Kun; Jung, Joo-Young; Suk Suh, Tae, E-mail: suhsanta@catholic.ac.kr

Purpose: The purpose of this research is to perform the fast reconstruction of a prompt gamma ray image using a graphics processing unit (GPU) computation from boron neutron capture therapy (BNCT) simulations. Methods: To evaluate the accuracy of the reconstructed image, a phantom including four boron uptake regions (BURs) was used in the simulation. After the Monte Carlo simulation of the BNCT, the modified ordered subset expectation maximization reconstruction algorithm using the GPU computation was used to reconstruct the images with fewer projections. The computation times for image reconstruction were compared between the GPU and the central processing unit (CPU).more » Also, the accuracy of the reconstructed image was evaluated by a receiver operating characteristic (ROC) curve analysis. Results: The image reconstruction time using the GPU was 196 times faster than the conventional reconstruction time using the CPU. For the four BURs, the area under curve values from the ROC curve were 0.6726 (A-region), 0.6890 (B-region), 0.7384 (C-region), and 0.8009 (D-region). Conclusions: The tomographic image using the prompt gamma ray event from the BNCT simulation was acquired using the GPU computation in order to perform a fast reconstruction during treatment. The authors verified the feasibility of the prompt gamma ray image reconstruction using the GPU computation for BNCT simulations.« less

TU-FG-BRB-07: GPU-Based Prompt Gamma Ray Imaging From Boron Neutron Capture Therapy

DOE Office of Scientific and Technical Information (OSTI.GOV)

Kim, S; Suh, T; Yoon, D

Purpose: The purpose of this research is to perform the fast reconstruction of a prompt gamma ray image using a graphics processing unit (GPU) computation from boron neutron capture therapy (BNCT) simulations. Methods: To evaluate the accuracy of the reconstructed image, a phantom including four boron uptake regions (BURs) was used in the simulation. After the Monte Carlo simulation of the BNCT, the modified ordered subset expectation maximization reconstruction algorithm using the GPU computation was used to reconstruct the images with fewer projections. The computation times for image reconstruction were compared between the GPU and the central processing unit (CPU).more » Also, the accuracy of the reconstructed image was evaluated by a receiver operating characteristic (ROC) curve analysis. Results: The image reconstruction time using the GPU was 196 times faster than the conventional reconstruction time using the CPU. For the four BURs, the area under curve values from the ROC curve were 0.6726 (A-region), 0.6890 (B-region), 0.7384 (C-region), and 0.8009 (D-region). Conclusion: The tomographic image using the prompt gamma ray event from the BNCT simulation was acquired using the GPU computation in order to perform a fast reconstruction during treatment. The authors verified the feasibility of the prompt gamma ray reconstruction using the GPU computation for BNCT simulations.« less

The CUBLAS and CULA based GPU acceleration of adaptive finite element framework for bioluminescence tomography.

PubMed

Zhang, Bo; Yang, Xiang; Yang, Fei; Yang, Xin; Qin, Chenghu; Han, Dong; Ma, Xibo; Liu, Kai; Tian, Jie

2010-09-13

In molecular imaging (MI), especially the optical molecular imaging, bioluminescence tomography (BLT) emerges as an effective imaging modality for small animal imaging. The finite element methods (FEMs), especially the adaptive finite element (AFE) framework, play an important role in BLT. The processing speed of the FEMs and the AFE framework still needs to be improved, although the multi-thread CPU technology and the multi CPU technology have already been applied. In this paper, we for the first time introduce a new kind of acceleration technology to accelerate the AFE framework for BLT, using the graphics processing unit (GPU). Besides the processing speed, the GPU technology can get a balance between the cost and performance. The CUBLAS and CULA are two main important and powerful libraries for programming on NVIDIA GPUs. With the help of CUBLAS and CULA, it is easy to code on NVIDIA GPU and there is no need to worry about the details about the hardware environment of a specific GPU. The numerical experiments are designed to show the necessity, effect and application of the proposed CUBLAS and CULA based GPU acceleration. From the results of the experiments, we can reach the conclusion that the proposed CUBLAS and CULA based GPU acceleration method can improve the processing speed of the AFE framework very much while getting a balance between cost and performance.

GPULife

DOE Office of Scientific and Technical Information (OSTI.GOV)

Kelly, Priscilla N.

2016-08-12

The code runs the Game of Life among several processors. Each processor uses CUDA to set up the grid's buffer on the GPU, and that buffer is fed to other GPU languages to apply the rules of the game of life. Only the halo is copied off the buffer and exchanged using MPI. This code looks at the interoperability of GPU languages among current platforms.

Rubus: A compiler for seamless and extensible parallelism.

PubMed

Adnan, Muhammad; Aslam, Faisal; Nawaz, Zubair; Sarwar, Syed Mansoor

2017-01-01

Nowadays, a typical processor may have multiple processing cores on a single chip. Furthermore, a special purpose processing unit called Graphic Processing Unit (GPU), originally designed for 2D/3D games, is now available for general purpose use in computers and mobile devices. However, the traditional programming languages which were designed to work with machines having single core CPUs, cannot utilize the parallelism available on multi-core processors efficiently. Therefore, to exploit the extraordinary processing power of multi-core processors, researchers are working on new tools and techniques to facilitate parallel programming. To this end, languages like CUDA and OpenCL have been introduced, which can be used to write code with parallelism. The main shortcoming of these languages is that programmer needs to specify all the complex details manually in order to parallelize the code across multiple cores. Therefore, the code written in these languages is difficult to understand, debug and maintain. Furthermore, to parallelize legacy code can require rewriting a significant portion of code in CUDA or OpenCL, which can consume significant time and resources. Thus, the amount of parallelism achieved is proportional to the skills of the programmer and the time spent in code optimizations. This paper proposes a new open source compiler, Rubus, to achieve seamless parallelism. The Rubus compiler relieves the programmer from manually specifying the low-level details. It analyses and transforms a sequential program into a parallel program automatically, without any user intervention. This achieves massive speedup and better utilization of the underlying hardware without a programmer's expertise in parallel programming. For five different benchmarks, on average a speedup of 34.54 times has been achieved by Rubus as compared to Java on a basic GPU having only 96 cores. Whereas, for a matrix multiplication benchmark the average execution speedup of 84 times has been achieved by Rubus on the same GPU. Moreover, Rubus achieves this performance without drastically increasing the memory footprint of a program.

Rubus: A compiler for seamless and extensible parallelism

PubMed Central

Adnan, Muhammad; Aslam, Faisal; Sarwar, Syed Mansoor

2017-01-01

Nowadays, a typical processor may have multiple processing cores on a single chip. Furthermore, a special purpose processing unit called Graphic Processing Unit (GPU), originally designed for 2D/3D games, is now available for general purpose use in computers and mobile devices. However, the traditional programming languages which were designed to work with machines having single core CPUs, cannot utilize the parallelism available on multi-core processors efficiently. Therefore, to exploit the extraordinary processing power of multi-core processors, researchers are working on new tools and techniques to facilitate parallel programming. To this end, languages like CUDA and OpenCL have been introduced, which can be used to write code with parallelism. The main shortcoming of these languages is that programmer needs to specify all the complex details manually in order to parallelize the code across multiple cores. Therefore, the code written in these languages is difficult to understand, debug and maintain. Furthermore, to parallelize legacy code can require rewriting a significant portion of code in CUDA or OpenCL, which can consume significant time and resources. Thus, the amount of parallelism achieved is proportional to the skills of the programmer and the time spent in code optimizations. This paper proposes a new open source compiler, Rubus, to achieve seamless parallelism. The Rubus compiler relieves the programmer from manually specifying the low-level details. It analyses and transforms a sequential program into a parallel program automatically, without any user intervention. This achieves massive speedup and better utilization of the underlying hardware without a programmer’s expertise in parallel programming. For five different benchmarks, on average a speedup of 34.54 times has been achieved by Rubus as compared to Java on a basic GPU having only 96 cores. Whereas, for a matrix multiplication benchmark the average execution speedup of 84 times has been achieved by Rubus on the same GPU. Moreover, Rubus achieves this performance without drastically increasing the memory footprint of a program. PMID:29211758

Democratic population decisions result in robust policy-gradient learning: a parametric study with GPU simulations.

PubMed

Richmond, Paul; Buesing, Lars; Giugliano, Michele; Vasilaki, Eleni

2011-05-04

High performance computing on the Graphics Processing Unit (GPU) is an emerging field driven by the promise of high computational power at a low cost. However, GPU programming is a non-trivial task and moreover architectural limitations raise the question of whether investing effort in this direction may be worthwhile. In this work, we use GPU programming to simulate a two-layer network of Integrate-and-Fire neurons with varying degrees of recurrent connectivity and investigate its ability to learn a simplified navigation task using a policy-gradient learning rule stemming from Reinforcement Learning. The purpose of this paper is twofold. First, we want to support the use of GPUs in the field of Computational Neuroscience. Second, using GPU computing power, we investigate the conditions under which the said architecture and learning rule demonstrate best performance. Our work indicates that networks featuring strong Mexican-Hat-shaped recurrent connections in the top layer, where decision making is governed by the formation of a stable activity bump in the neural population (a "non-democratic" mechanism), achieve mediocre learning results at best. In absence of recurrent connections, where all neurons "vote" independently ("democratic") for a decision via population vector readout, the task is generally learned better and more robustly. Our study would have been extremely difficult on a desktop computer without the use of GPU programming. We present the routines developed for this purpose and show that a speed improvement of 5x up to 42x is provided versus optimised Python code. The higher speed is achieved when we exploit the parallelism of the GPU in the search of learning parameters. This suggests that efficient GPU programming can significantly reduce the time needed for simulating networks of spiking neurons, particularly when multiple parameter configurations are investigated.

CUDASW++ 3.0: accelerating Smith-Waterman protein database search by coupling CPU and GPU SIMD instructions.

PubMed

Liu, Yongchao; Wirawan, Adrianto; Schmidt, Bertil

2013-04-04

The maximal sensitivity for local alignments makes the Smith-Waterman algorithm a popular choice for protein sequence database search based on pairwise alignment. However, the algorithm is compute-intensive due to a quadratic time complexity. Corresponding runtimes are further compounded by the rapid growth of sequence databases. We present CUDASW++ 3.0, a fast Smith-Waterman protein database search algorithm, which couples CPU and GPU SIMD instructions and carries out concurrent CPU and GPU computations. For the CPU computation, this algorithm employs SSE-based vector execution units as accelerators. For the GPU computation, we have investigated for the first time a GPU SIMD parallelization, which employs CUDA PTX SIMD video instructions to gain more data parallelism beyond the SIMT execution model. Moreover, sequence alignment workloads are automatically distributed over CPUs and GPUs based on their respective compute capabilities. Evaluation on the Swiss-Prot database shows that CUDASW++ 3.0 gains a performance improvement over CUDASW++ 2.0 up to 2.9 and 3.2, with a maximum performance of 119.0 and 185.6 GCUPS, on a single-GPU GeForce GTX 680 and a dual-GPU GeForce GTX 690 graphics card, respectively. In addition, our algorithm has demonstrated significant speedups over other top-performing tools: SWIPE and BLAST+. CUDASW++ 3.0 is written in CUDA C++ and PTX assembly languages, targeting GPUs based on the Kepler architecture. This algorithm obtains significant speedups over its predecessor: CUDASW++ 2.0, by benefiting from the use of CPU and GPU SIMD instructions as well as the concurrent execution on CPUs and GPUs. The source code and the simulated data are available at http://cudasw.sourceforge.net.

Revisiting Molecular Dynamics on a CPU/GPU system: Water Kernel and SHAKE Parallelization.

PubMed

Ruymgaart, A Peter; Elber, Ron

2012-11-13

We report Graphics Processing Unit (GPU) and Open-MP parallel implementations of water-specific force calculations and of bond constraints for use in Molecular Dynamics simulations. We focus on a typical laboratory computing-environment in which a CPU with a few cores is attached to a GPU. We discuss in detail the design of the code and we illustrate performance comparable to highly optimized codes such as GROMACS. Beside speed our code shows excellent energy conservation. Utilization of water-specific lists allows the efficient calculations of non-bonded interactions that include water molecules and results in a speed-up factor of more than 40 on the GPU compared to code optimized on a single CPU core for systems larger than 20,000 atoms. This is up four-fold from a factor of 10 reported in our initial GPU implementation that did not include a water-specific code. Another optimization is the implementation of constrained dynamics entirely on the GPU. The routine, which enforces constraints of all bonds, runs in parallel on multiple Open-MP cores or entirely on the GPU. It is based on Conjugate Gradient solution of the Lagrange multipliers (CG SHAKE). The GPU implementation is partially in double precision and requires no communication with the CPU during the execution of the SHAKE algorithm. The (parallel) implementation of SHAKE allows an increase of the time step to 2.0fs while maintaining excellent energy conservation. Interestingly, CG SHAKE is faster than the usual bond relaxation algorithm even on a single core if high accuracy is expected. The significant speedup of the optimized components transfers the computational bottleneck of the MD calculation to the reciprocal part of Particle Mesh Ewald (PME).

GPU-accelerated track reconstruction in the ALICE High Level Trigger

NASA Astrophysics Data System (ADS)

Rohr, David; Gorbunov, Sergey; Lindenstruth, Volker; ALICE Collaboration

2017-10-01

ALICE (A Large Heavy Ion Experiment) is one of the four major experiments at the Large Hadron Collider (LHC) at CERN. The High Level Trigger (HLT) is an online compute farm which reconstructs events measured by the ALICE detector in real-time. The most compute-intensive part is the reconstruction of particle trajectories called tracking and the most important detector for tracking is the Time Projection Chamber (TPC). The HLT uses a GPU-accelerated algorithm for TPC tracking that is based on the Cellular Automaton principle and on the Kalman filter. The GPU tracking has been running in 24/7 operation since 2012 in LHC Run 1 and 2. In order to better leverage the potential of the GPUs, and speed up the overall HLT reconstruction, we plan to bring more reconstruction steps (e.g. the tracking for other detectors) onto the GPUs. There are several tasks running so far on the CPU that could benefit from cooperation with the tracking, which is hardly feasible at the moment due to the delay of the PCI Express transfers. Moving more steps onto the GPU, and processing them on the GPU at once, will reduce PCI Express transfers and free up CPU resources. On top of that, modern GPUs and GPU programming APIs provide new features which are not yet exploited by the TPC tracking. We present our new developments for GPU reconstruction, both with a focus on the online reconstruction on GPU for the online offline computing upgrade in ALICE during LHC Run 3, and also taking into account how the current HLT in Run 2 can profit from these improvements.

Semiempirical Quantum Chemical Calculations Accelerated on a Hybrid Multicore CPU-GPU Computing Platform.

PubMed

Wu, Xin; Koslowski, Axel; Thiel, Walter

2012-07-10

In this work, we demonstrate that semiempirical quantum chemical calculations can be accelerated significantly by leveraging the graphics processing unit (GPU) as a coprocessor on a hybrid multicore CPU-GPU computing platform. Semiempirical calculations using the MNDO, AM1, PM3, OM1, OM2, and OM3 model Hamiltonians were systematically profiled for three types of test systems (fullerenes, water clusters, and solvated crambin) to identify the most time-consuming sections of the code. The corresponding routines were ported to the GPU and optimized employing both existing library functions and a GPU kernel that carries out a sequence of noniterative Jacobi transformations during pseudodiagonalization. The overall computation times for single-point energy calculations and geometry optimizations of large molecules were reduced by one order of magnitude for all methods, as compared to runs on a single CPU core.

GPU-Powered Coherent Beamforming

NASA Astrophysics Data System (ADS)

Magro, A.; Adami, K. Zarb; Hickish, J.

2015-03-01

Graphics processing units (GPU)-based beamforming is a relatively unexplored area in radio astronomy, possibly due to the assumption that any such system will be severely limited by the PCIe bandwidth required to transfer data to the GPU. We have developed a CUDA-based GPU implementation of a coherent beamformer, specifically designed and optimized for deployment at the BEST-2 array which can generate an arbitrary number of synthesized beams for a wide range of parameters. It achieves ˜1.3 TFLOPs on an NVIDIA Tesla K20, approximately 10x faster than an optimized, multithreaded CPU implementation. This kernel has been integrated into two real-time, GPU-based time-domain software pipelines deployed at the BEST-2 array in Medicina: a standalone beamforming pipeline and a transient detection pipeline. We present performance benchmarks for the beamforming kernel as well as the transient detection pipeline with beamforming capabilities as well as results of test observation.

Improving GPU-accelerated adaptive IDW interpolation algorithm using fast kNN search.

PubMed

Mei, Gang; Xu, Nengxiong; Xu, Liangliang

2016-01-01

This paper presents an efficient parallel Adaptive Inverse Distance Weighting (AIDW) interpolation algorithm on modern Graphics Processing Unit (GPU). The presented algorithm is an improvement of our previous GPU-accelerated AIDW algorithm by adopting fast k-nearest neighbors (kNN) search. In AIDW, it needs to find several nearest neighboring data points for each interpolated point to adaptively determine the power parameter; and then the desired prediction value of the interpolated point is obtained by weighted interpolating using the power parameter. In this work, we develop a fast kNN search approach based on the space-partitioning data structure, even grid, to improve the previous GPU-accelerated AIDW algorithm. The improved algorithm is composed of the stages of kNN search and weighted interpolating. To evaluate the performance of the improved algorithm, we perform five groups of experimental tests. The experimental results indicate: (1) the improved algorithm can achieve a speedup of up to 1017 over the corresponding serial algorithm; (2) the improved algorithm is at least two times faster than our previous GPU-accelerated AIDW algorithm; and (3) the utilization of fast kNN search can significantly improve the computational efficiency of the entire GPU-accelerated AIDW algorithm.

Compiler-based code generation and autotuning for geometric multigrid on GPU-accelerated supercomputers

DOE PAGES

Basu, Protonu; Williams, Samuel; Van Straalen, Brian; ...

2017-04-05

GPUs, with their high bandwidths and computational capabilities are an increasingly popular target for scientific computing. Unfortunately, to date, harnessing the power of the GPU has required use of a GPU-specific programming model like CUDA, OpenCL, or OpenACC. Thus, in order to deliver portability across CPU-based and GPU-accelerated supercomputers, programmers are forced to write and maintain two versions of their applications or frameworks. In this paper, we explore the use of a compiler-based autotuning framework based on CUDA-CHiLL to deliver not only portability, but also performance portability across CPU- and GPU-accelerated platforms for the geometric multigrid linear solvers found inmore » many scientific applications. We also show that with autotuning we can attain near Roofline (a performance bound for a computation and target architecture) performance across the key operations in the miniGMG benchmark for both CPU- and GPU-based architectures as well as for a multiple stencil discretizations and smoothers. We show that our technology is readily interoperable with MPI resulting in performance at scale equal to that obtained via hand-optimized MPI+CUDA implementation.« less

GPU accelerated simulations of 3D deterministic particle transport using discrete ordinates method

NASA Astrophysics Data System (ADS)

Gong, Chunye; Liu, Jie; Chi, Lihua; Huang, Haowei; Fang, Jingyue; Gong, Zhenghu

2011-07-01

Graphics Processing Unit (GPU), originally developed for real-time, high-definition 3D graphics in computer games, now provides great faculty in solving scientific applications. The basis of particle transport simulation is the time-dependent, multi-group, inhomogeneous Boltzmann transport equation. The numerical solution to the Boltzmann equation involves the discrete ordinates ( Sn) method and the procedure of source iteration. In this paper, we present a GPU accelerated simulation of one energy group time-independent deterministic discrete ordinates particle transport in 3D Cartesian geometry (Sweep3D). The performance of the GPU simulations are reported with the simulations of vacuum boundary condition. The discussion of the relative advantages and disadvantages of the GPU implementation, the simulation on multi GPUs, the programming effort and code portability are also reported. The results show that the overall performance speedup of one NVIDIA Tesla M2050 GPU ranges from 2.56 compared with one Intel Xeon X5670 chip to 8.14 compared with one Intel Core Q6600 chip for no flux fixup. The simulation with flux fixup on one M2050 is 1.23 times faster than on one X5670.

Parallel Optimization of 3D Cardiac Electrophysiological Model Using GPU

PubMed Central

Xia, Yong; Zhang, Henggui

2015-01-01

Large-scale 3D virtual heart model simulations are highly demanding in computational resources. This imposes a big challenge to the traditional computation resources based on CPU environment, which already cannot meet the requirement of the whole computation demands or are not easily available due to expensive costs. GPU as a parallel computing environment therefore provides an alternative to solve the large-scale computational problems of whole heart modeling. In this study, using a 3D sheep atrial model as a test bed, we developed a GPU-based simulation algorithm to simulate the conduction of electrical excitation waves in the 3D atria. In the GPU algorithm, a multicellular tissue model was split into two components: one is the single cell model (ordinary differential equation) and the other is the diffusion term of the monodomain model (partial differential equation). Such a decoupling enabled realization of the GPU parallel algorithm. Furthermore, several optimization strategies were proposed based on the features of the virtual heart model, which enabled a 200-fold speedup as compared to a CPU implementation. In conclusion, an optimized GPU algorithm has been developed that provides an economic and powerful platform for 3D whole heart simulations. PMID:26581957

High Performance GPU-Based Fourier Volume Rendering.

PubMed

Abdellah, Marwan; Eldeib, Ayman; Sharawi, Amr

2015-01-01

Fourier volume rendering (FVR) is a significant visualization technique that has been used widely in digital radiography. As a result of its (N (2)log⁡N) time complexity, it provides a faster alternative to spatial domain volume rendering algorithms that are (N (3)) computationally complex. Relying on the Fourier projection-slice theorem, this technique operates on the spectral representation of a 3D volume instead of processing its spatial representation to generate attenuation-only projections that look like X-ray radiographs. Due to the rapid evolution of its underlying architecture, the graphics processing unit (GPU) became an attractive competent platform that can deliver giant computational raw power compared to the central processing unit (CPU) on a per-dollar-basis. The introduction of the compute unified device architecture (CUDA) technology enables embarrassingly-parallel algorithms to run efficiently on CUDA-capable GPU architectures. In this work, a high performance GPU-accelerated implementation of the FVR pipeline on CUDA-enabled GPUs is presented. This proposed implementation can achieve a speed-up of 117x compared to a single-threaded hybrid implementation that uses the CPU and GPU together by taking advantage of executing the rendering pipeline entirely on recent GPU architectures.

Parallel Optimization of 3D Cardiac Electrophysiological Model Using GPU.

PubMed

Xia, Yong; Wang, Kuanquan; Zhang, Henggui

2015-01-01

Large-scale 3D virtual heart model simulations are highly demanding in computational resources. This imposes a big challenge to the traditional computation resources based on CPU environment, which already cannot meet the requirement of the whole computation demands or are not easily available due to expensive costs. GPU as a parallel computing environment therefore provides an alternative to solve the large-scale computational problems of whole heart modeling. In this study, using a 3D sheep atrial model as a test bed, we developed a GPU-based simulation algorithm to simulate the conduction of electrical excitation waves in the 3D atria. In the GPU algorithm, a multicellular tissue model was split into two components: one is the single cell model (ordinary differential equation) and the other is the diffusion term of the monodomain model (partial differential equation). Such a decoupling enabled realization of the GPU parallel algorithm. Furthermore, several optimization strategies were proposed based on the features of the virtual heart model, which enabled a 200-fold speedup as compared to a CPU implementation. In conclusion, an optimized GPU algorithm has been developed that provides an economic and powerful platform for 3D whole heart simulations.

Compiler-based code generation and autotuning for geometric multigrid on GPU-accelerated supercomputers

DOE Office of Scientific and Technical Information (OSTI.GOV)

Basu, Protonu; Williams, Samuel; Van Straalen, Brian

GPUs, with their high bandwidths and computational capabilities are an increasingly popular target for scientific computing. Unfortunately, to date, harnessing the power of the GPU has required use of a GPU-specific programming model like CUDA, OpenCL, or OpenACC. Thus, in order to deliver portability across CPU-based and GPU-accelerated supercomputers, programmers are forced to write and maintain two versions of their applications or frameworks. In this paper, we explore the use of a compiler-based autotuning framework based on CUDA-CHiLL to deliver not only portability, but also performance portability across CPU- and GPU-accelerated platforms for the geometric multigrid linear solvers found inmore » many scientific applications. We also show that with autotuning we can attain near Roofline (a performance bound for a computation and target architecture) performance across the key operations in the miniGMG benchmark for both CPU- and GPU-based architectures as well as for a multiple stencil discretizations and smoothers. We show that our technology is readily interoperable with MPI resulting in performance at scale equal to that obtained via hand-optimized MPI+CUDA implementation.« less

Transient Global Amnesia

MedlinePlus

... global amnesia is a sudden, temporary episode of memory loss that can't be attributed to a ... know well. But that doesn't make your memory loss less disturbing. Fortunately, transient global amnesia is ...

Global neural pattern similarity as a common basis for categorization and recognition memory.

PubMed

Davis, Tyler; Xue, Gui; Love, Bradley C; Preston, Alison R; Poldrack, Russell A

2014-05-28

Familiarity, or memory strength, is a central construct in models of cognition. In previous categorization and long-term memory research, correlations have been found between psychological measures of memory strength and activation in the medial temporal lobes (MTLs), which suggests a common neural locus for memory strength. However, activation alone is insufficient for determining whether the same mechanisms underlie neural function across domains. Guided by mathematical models of categorization and long-term memory, we develop a theory and a method to test whether memory strength arises from the global similarity among neural representations. In human subjects, we find significant correlations between global similarity among activation patterns in the MTLs and both subsequent memory confidence in a recognition memory task and model-based measures of memory strength in a category learning task. Our work bridges formal cognitive theories and neuroscientific models by illustrating that the same global similarity computations underlie processing in multiple cognitive domains. Moreover, by establishing a link between neural similarity and psychological memory strength, our findings suggest that there may be an isomorphism between psychological and neural representational spaces that can be exploited to test cognitive theories at both the neural and behavioral levels. Copyright © 2014 the authors 0270-6474/14/347472-13$15.00/0.

Large Scale Document Inversion using a Multi-threaded Computing System

PubMed Central

Jung, Sungbo; Chang, Dar-Jen; Park, Juw Won

2018-01-01

Current microprocessor architecture is moving towards multi-core/multi-threaded systems. This trend has led to a surge of interest in using multi-threaded computing devices, such as the Graphics Processing Unit (GPU), for general purpose computing. We can utilize the GPU in computation as a massive parallel coprocessor because the GPU consists of multiple cores. The GPU is also an affordable, attractive, and user-programmable commodity. Nowadays a lot of information has been flooded into the digital domain around the world. Huge volume of data, such as digital libraries, social networking services, e-commerce product data, and reviews, etc., is produced or collected every moment with dramatic growth in size. Although the inverted index is a useful data structure that can be used for full text searches or document retrieval, a large number of documents will require a tremendous amount of time to create the index. The performance of document inversion can be improved by multi-thread or multi-core GPU. Our approach is to implement a linear-time, hash-based, single program multiple data (SPMD), document inversion algorithm on the NVIDIA GPU/CUDA programming platform utilizing the huge computational power of the GPU, to develop high performance solutions for document indexing. Our proposed parallel document inversion system shows 2-3 times faster performance than a sequential system on two different test datasets from PubMed abstract and e-commerce product reviews. CCS Concepts •Information systems➝Information retrieval • Computing methodologies➝Massively parallel and high-performance simulations. PMID:29861701

Novel hybrid GPU-CPU implementation of parallelized Monte Carlo parametric expectation maximization estimation method for population pharmacokinetic data analysis.

PubMed

Ng, C M

2013-10-01

The development of a population PK/PD model, an essential component for model-based drug development, is both time- and labor-intensive. A graphical-processing unit (GPU) computing technology has been proposed and used to accelerate many scientific computations. The objective of this study was to develop a hybrid GPU-CPU implementation of parallelized Monte Carlo parametric expectation maximization (MCPEM) estimation algorithm for population PK data analysis. A hybrid GPU-CPU implementation of the MCPEM algorithm (MCPEMGPU) and identical algorithm that is designed for the single CPU (MCPEMCPU) were developed using MATLAB in a single computer equipped with dual Xeon 6-Core E5690 CPU and a NVIDIA Tesla C2070 GPU parallel computing card that contained 448 stream processors. Two different PK models with rich/sparse sampling design schemes were used to simulate population data in assessing the performance of MCPEMCPU and MCPEMGPU. Results were analyzed by comparing the parameter estimation and model computation times. Speedup factor was used to assess the relative benefit of parallelized MCPEMGPU over MCPEMCPU in shortening model computation time. The MCPEMGPU consistently achieved shorter computation time than the MCPEMCPU and can offer more than 48-fold speedup using a single GPU card. The novel hybrid GPU-CPU implementation of parallelized MCPEM algorithm developed in this study holds a great promise in serving as the core for the next-generation of modeling software for population PK/PD analysis.

Large Scale Document Inversion using a Multi-threaded Computing System.

PubMed

Jung, Sungbo; Chang, Dar-Jen; Park, Juw Won

2017-06-01

Current microprocessor architecture is moving towards multi-core/multi-threaded systems. This trend has led to a surge of interest in using multi-threaded computing devices, such as the Graphics Processing Unit (GPU), for general purpose computing. We can utilize the GPU in computation as a massive parallel coprocessor because the GPU consists of multiple cores. The GPU is also an affordable, attractive, and user-programmable commodity. Nowadays a lot of information has been flooded into the digital domain around the world. Huge volume of data, such as digital libraries, social networking services, e-commerce product data, and reviews, etc., is produced or collected every moment with dramatic growth in size. Although the inverted index is a useful data structure that can be used for full text searches or document retrieval, a large number of documents will require a tremendous amount of time to create the index. The performance of document inversion can be improved by multi-thread or multi-core GPU. Our approach is to implement a linear-time, hash-based, single program multiple data (SPMD), document inversion algorithm on the NVIDIA GPU/CUDA programming platform utilizing the huge computational power of the GPU, to develop high performance solutions for document indexing. Our proposed parallel document inversion system shows 2-3 times faster performance than a sequential system on two different test datasets from PubMed abstract and e-commerce product reviews. •Information systems➝Information retrieval • Computing methodologies➝Massively parallel and high-performance simulations.

3D superwide-angle one-way propagator and its application in seismic modeling and imaging

NASA Astrophysics Data System (ADS)

Jia, Xiaofeng; Jiang, Yunong; Wu, Ru-Shan

2018-07-01

Traditional one-way wave-equation based propagators have been widely used in past decades. Comparing to two-way propagators, one-way methods have higher efficiency and lower memory demands. These two features are especially important in solving large-scale 3D problems. However, regular one-way propagators cannot simulate waves that propagate in large angles within 90° because of their inherent wide angle limitation. Traditional one-way can only propagate along the determined direction (e.g., z-direction), so simulation of turning waves is beyond the ability of one-way methods. We develop 3D superwide-angle one-way propagator to overcome angle limitation and to simulate turning waves with superwide-angle propagation angle (>90°) for modeling and imaging complex geological structures. Wavefields propagating along vertical and horizontal directions are combined using typical stacking scheme. A weight function related to the propagation angle is used for combining and updating wavefields in each propagating step. In the implementation, we use graphics processing units (GPU) to accelerate the process. Typical workflow is designed to exploit the advantages of GPU architecture. Numerical examples show that the method achieves higher accuracy in modeling and imaging steep structures than regular one-way propagators. Actually, superwide-angle one-way propagator can be applied based on any one-way method to improve the effects of seismic modeling and imaging.

GPU applications for data processing

DOE Office of Scientific and Technical Information (OSTI.GOV)

Vladymyrov, Mykhailo, E-mail: mykhailo.vladymyrov@cern.ch; Aleksandrov, Andrey; INFN sezione di Napoli, I-80125 Napoli

2015-12-31

Modern experiments that use nuclear photoemulsion imply fast and efficient data acquisition from the emulsion can be performed. The new approaches in developing scanning systems require real-time processing of large amount of data. Methods that use Graphical Processing Unit (GPU) computing power for emulsion data processing are presented here. It is shown how the GPU-accelerated emulsion processing helped us to rise the scanning speed by factor of nine.

TH-A-18C-09: Ultra-Fast Monte Carlo Simulation for Cone Beam CT Imaging of Brain Trauma

DOE Office of Scientific and Technical Information (OSTI.GOV)

Sisniega, A; Zbijewski, W; Stayman, J

Purpose: Application of cone-beam CT (CBCT) to low-contrast soft tissue imaging, such as in detection of traumatic brain injury, is challenged by high levels of scatter. A fast, accurate scatter correction method based on Monte Carlo (MC) estimation is developed for application in high-quality CBCT imaging of acute brain injury. Methods: The correction involves MC scatter estimation executed on an NVIDIA GTX 780 GPU (MC-GPU), with baseline simulation speed of ~1e7 photons/sec. MC-GPU is accelerated by a novel, GPU-optimized implementation of variance reduction (VR) techniques (forced detection and photon splitting). The number of simulated tracks and projections is reduced formore » additional speed-up. Residual noise is removed and the missing scatter projections are estimated via kernel smoothing (KS) in projection plane and across gantry angles. The method is assessed using CBCT images of a head phantom presenting a realistic simulation of fresh intracranial hemorrhage (100 kVp, 180 mAs, 720 projections, source-detector distance 700 mm, source-axis distance 480 mm). Results: For a fixed run-time of ~1 sec/projection, GPU-optimized VR reduces the noise in MC-GPU scatter estimates by a factor of 4. For scatter correction, MC-GPU with VR is executed with 4-fold angular downsampling and 1e5 photons/projection, yielding 3.5 minute run-time per scan, and de-noised with optimized KS. Corrected CBCT images demonstrate uniformity improvement of 18 HU and contrast improvement of 26 HU compared to no correction, and a 52% increase in contrast-tonoise ratio in simulated hemorrhage compared to “oracle” constant fraction correction. Conclusion: Acceleration of MC-GPU achieved through GPU-optimized variance reduction and kernel smoothing yields an efficient (<5 min/scan) and accurate scatter correction that does not rely on additional hardware or simplifying assumptions about the scatter distribution. The method is undergoing implementation in a novel CBCT dedicated to brain trauma imaging at the point of care in sports and military applications. Research grant from Carestream Health. JY is an employee of Carestream Health.« less

Distributed GPU Computing in GIScience

NASA Astrophysics Data System (ADS)

Jiang, Y.; Yang, C.; Huang, Q.; Li, J.; Sun, M.

2013-12-01

Geoscientists strived to discover potential principles and patterns hidden inside ever-growing Big Data for scientific discoveries. To better achieve this objective, more capable computing resources are required to process, analyze and visualize Big Data (Ferreira et al., 2003; Li et al., 2013). Current CPU-based computing techniques cannot promptly meet the computing challenges caused by increasing amount of datasets from different domains, such as social media, earth observation, environmental sensing (Li et al., 2013). Meanwhile CPU-based computing resources structured as cluster or supercomputer is costly. In the past several years with GPU-based technology matured in both the capability and performance, GPU-based computing has emerged as a new computing paradigm. Compare to traditional computing microprocessor, the modern GPU, as a compelling alternative microprocessor, has outstanding high parallel processing capability with cost-effectiveness and efficiency(Owens et al., 2008), although it is initially designed for graphical rendering in visualization pipe. This presentation reports a distributed GPU computing framework for integrating GPU-based computing within distributed environment. Within this framework, 1) for each single computer, computing resources of both GPU-based and CPU-based can be fully utilized to improve the performance of visualizing and processing Big Data; 2) within a network environment, a variety of computers can be used to build up a virtual super computer to support CPU-based and GPU-based computing in distributed computing environment; 3) GPUs, as a specific graphic targeted device, are used to greatly improve the rendering efficiency in distributed geo-visualization, especially for 3D/4D visualization. Key words: Geovisualization, GIScience, Spatiotemporal Studies Reference : 1. Ferreira de Oliveira, M. C., & Levkowitz, H. (2003). From visual data exploration to visual data mining: A survey. Visualization and Computer Graphics, IEEE Transactions on, 9(3), 378-394. 2. Li, J., Jiang, Y., Yang, C., Huang, Q., & Rice, M. (2013). Visualizing 3D/4D Environmental Data Using Many-core Graphics Processing Units (GPUs) and Multi-core Central Processing Units (CPUs). Computers & Geosciences, 59(9), 78-89. 3. Owens, J. D., Houston, M., Luebke, D., Green, S., Stone, J. E., & Phillips, J. C. (2008). GPU computing. Proceedings of the IEEE, 96(5), 879-899.

DOE Office of Scientific and Technical Information (OSTI.GOV)

Ibrahim, Khaled Z.; Epifanovsky, Evgeny; Williams, Samuel W.

Coupled-cluster methods provide highly accurate models of molecular structure by explicit numerical calculation of tensors representing the correlation between electrons. These calculations are dominated by a sequence of tensor contractions, motivating the development of numerical libraries for such operations. While based on matrix-matrix multiplication, these libraries are specialized to exploit symmetries in the molecular structure and in electronic interactions, and thus reduce the size of the tensor representation and the complexity of contractions. The resulting algorithms are irregular and their parallelization has been previously achieved via the use of dynamic scheduling or specialized data decompositions. We introduce our efforts tomore » extend the Libtensor framework to work in the distributed memory environment in a scalable and energy efficient manner. We achieve up to 240 speedup compared with the best optimized shared memory implementation. We attain scalability to hundreds of thousands of compute cores on three distributed-memory architectures, (Cray XC30&XC40, BlueGene/Q), and on a heterogeneous GPU-CPU system (Cray XK7). As the bottlenecks shift from being compute-bound DGEMM's to communication-bound collectives as the size of the molecular system scales, we adopt two radically different parallelization approaches for handling load-imbalance. Nevertheless, we preserve a uni ed interface to both programming models to maintain the productivity of computational quantum chemists.« less

Global Adjoint Tomography: Next-Generation Models

NASA Astrophysics Data System (ADS)

Bozdag, Ebru; Lefebvre, Matthieu; Lei, Wenjie; Orsvuran, Ridvan; Peter, Daniel; Ruan, Youyi; Smith, James; Komatitsch, Dimitri; Tromp, Jeroen

2017-04-01

The first-generation global adjoint tomography model GLAD-M15 (Bozdag et al. 2016) is the result of 15 conjugate-gradient iterations based on GPU-accelerated spectral-element simulations of 3D wave propagation and Fréchet kernels. For simplicity, GLAD-M15 was constructed as an elastic model with transverse isotropy confined to the upper mantle. However, Earth's mantle and crust show significant evidence of anisotropy as a result of its composition and deformation. There may be different sources of seismic anisotropy affecting both body and surface waves. As a first attempt, we initially tackle with surface-wave anisotropy and proceed iterations using the same 253 earthquake data set used in GLAD-M15 with an emphasize on upper-mantle. Furthermore, we explore new misfits, such as double-difference measurements (Yuan et al. 2016), to better deal with the possible artifacts of the uneven distribution of seismic stations globally and minimize source uncertainties in structural inversions. We will present our observations with the initial results of azimuthally anisotropic inversions and also discuss the next generation global models with various parametrizations. Meanwhile our goal is to use all available seismic data in imaging. This however requires a solid framework to perform iterative adjoint tomography workflows with big data on supercomputers. We will talk about developments in adjoint tomography workflow from the need of defining new seismic and computational data formats (e.g., ASDF by Krischer et al. 2016, ADIOS by Liu et al. 2011) to developing new pre- and post-processing tools together with experimenting workflow management tools, such as Pegasus (Deelman et al. 2015). All our simulations are performed on Oak Ridge National Laboratory's Cray XK7 "Titan" system. Our ultimate aim is to get ready to harness ORNL's next-generation supercomputer "Summit", an IBM with Power-9 CPUs and NVIDIA Volta GPU accelerators, to be ready by 2018 which will enable us to reduce the shortest period in our global simulations from 17 s to 9 s, and exascale systems will reduce this further to just a few seconds.

The Research and Test of Fast Radio Burst Real-time Search Algorithm Based on GPU Acceleration

NASA Astrophysics Data System (ADS)

Wang, J.; Chen, M. Z.; Pei, X.; Wang, Z. Q.

2017-03-01

In order to satisfy the research needs of Nanshan 25 m radio telescope of Xinjiang Astronomical Observatory (XAO) and study the key technology of the planned QiTai radio Telescope (QTT), the receiver group of XAO studied the GPU (Graphics Processing Unit) based real-time FRB searching algorithm which developed from the original FRB searching algorithm based on CPU (Central Processing Unit), and built the FRB real-time searching system. The comparison of the GPU system and the CPU system shows that: on the basis of ensuring the accuracy of the search, the speed of the GPU accelerated algorithm is improved by 35-45 times compared with the CPU algorithm.

GPU-accelerated computation of electron transfer.

PubMed

Höfinger, Siegfried; Acocella, Angela; Pop, Sergiu C; Narumi, Tetsu; Yasuoka, Kenji; Beu, Titus; Zerbetto, Francesco

2012-11-05

Electron transfer is a fundamental process that can be studied with the help of computer simulation. The underlying quantum mechanical description renders the problem a computationally intensive application. In this study, we probe the graphics processing unit (GPU) for suitability to this type of problem. Time-critical components are identified via profiling of an existing implementation and several different variants are tested involving the GPU at increasing levels of abstraction. A publicly available library supporting basic linear algebra operations on the GPU turns out to accelerate the computation approximately 50-fold with minor dependence on actual problem size. The performance gain does not compromise numerical accuracy and is of significant value for practical purposes. Copyright © 2012 Wiley Periodicals, Inc.

GPU Accelerated Prognostics

NASA Technical Reports Server (NTRS)

Gorospe, George E., Jr.; Daigle, Matthew J.; Sankararaman, Shankar; Kulkarni, Chetan S.; Ng, Eley

2017-01-01

Prognostic methods enable operators and maintainers to predict the future performance for critical systems. However, these methods can be computationally expensive and may need to be performed each time new information about the system becomes available. In light of these computational requirements, we have investigated the application of graphics processing units (GPUs) as a computational platform for real-time prognostics. Recent advances in GPU technology have reduced cost and increased the computational capability of these highly parallel processing units, making them more attractive for the deployment of prognostic software. We present a survey of model-based prognostic algorithms with considerations for leveraging the parallel architecture of the GPU and a case study of GPU-accelerated battery prognostics with computational performance results.

Maximal clique enumeration with data-parallel primitives

DOE Office of Scientific and Technical Information (OSTI.GOV)

Lessley, Brenton; Perciano, Talita; Mathai, Manish

The enumeration of all maximal cliques in an undirected graph is a fundamental problem arising in several research areas. We consider maximal clique enumeration on shared-memory, multi-core architectures and introduce an approach consisting entirely of data-parallel operations, in an effort to achieve efficient and portable performance across different architectures. We study the performance of the algorithm via experiments varying over benchmark graphs and architectures. Overall, we observe that our algorithm achieves up to a 33-time speedup and 9-time speedup over state-of-the-art distributed and serial algorithms, respectively, for graphs with higher ratios of maximal cliques to total cliques. Further, we attainmore » additional speedups on a GPU architecture, demonstrating the portable performance of our data-parallel design.« less

Optimizing a mobile robot control system using GPU acceleration

NASA Astrophysics Data System (ADS)

Tuck, Nat; McGuinness, Michael; Martin, Fred

2012-01-01

This paper describes our attempt to optimize a robot control program for the Intelligent Ground Vehicle Competition (IGVC) by running computationally intensive portions of the system on a commodity graphics processing unit (GPU). The IGVC Autonomous Challenge requires a control program that performs a number of different computationally intensive tasks ranging from computer vision to path planning. For the 2011 competition our Robot Operating System (ROS) based control system would not run comfortably on the multicore CPU on our custom robot platform. The process of profiling the ROS control program and selecting appropriate modules for porting to run on a GPU is described. A GPU-targeting compiler, Bacon, is used to speed up development and help optimize the ported modules. The impact of the ported modules on overall performance is discussed. We conclude that GPU optimization can free a significant amount of CPU resources with minimal effort for expensive user-written code, but that replacing heavily-optimized library functions is more difficult, and a much less efficient use of time.

Transportable GPU (General Processor Units) chip set technology for standard computer architectures

NASA Astrophysics Data System (ADS)

Fosdick, R. E.; Denison, H. C.

1982-11-01

The USAFR-developed GPU Chip Set has been utilized by Tracor to implement both USAF and Navy Standard 16-Bit Airborne Computer Architectures. Both configurations are currently being delivered into DOD full-scale development programs. Leadless Hermetic Chip Carrier packaging has facilitated implementation of both architectures on single 41/2 x 5 substrates. The CMOS and CMOS/SOS implementations of the GPU Chip Set have allowed both CPU implementations to use less than 3 watts of power each. Recent efforts by Tracor for USAF have included the definition of a next-generation GPU Chip Set that will retain the application-proven architecture of the current chip set while offering the added cost advantages of transportability across ISO-CMOS and CMOS/SOS processes and across numerous semiconductor manufacturers using a newly-defined set of common design rules. The Enhanced GPU Chip Set will increase speed by an approximate factor of 3 while significantly reducing chip counts and costs of standard CPU implementations.

A survey of GPU-based medical image computing techniques

PubMed Central

Shi, Lin; Liu, Wen; Zhang, Heye; Xie, Yongming

2012-01-01

Medical imaging currently plays a crucial role throughout the entire clinical applications from medical scientific research to diagnostics and treatment planning. However, medical imaging procedures are often computationally demanding due to the large three-dimensional (3D) medical datasets to process in practical clinical applications. With the rapidly enhancing performances of graphics processors, improved programming support, and excellent price-to-performance ratio, the graphics processing unit (GPU) has emerged as a competitive parallel computing platform for computationally expensive and demanding tasks in a wide range of medical image applications. The major purpose of this survey is to provide a comprehensive reference source for the starters or researchers involved in GPU-based medical image processing. Within this survey, the continuous advancement of GPU computing is reviewed and the existing traditional applications in three areas of medical image processing, namely, segmentation, registration and visualization, are surveyed. The potential advantages and associated challenges of current GPU-based medical imaging are also discussed to inspire future applications in medicine. PMID:23256080

The GPU implementation of micro - Doppler period estimation

NASA Astrophysics Data System (ADS)

Yang, Liyuan; Wang, Junling; Bi, Ran

2018-03-01

Aiming at the problem that the computational complexity and the deficiency of real-time of the wideband radar echo signal, a program is designed to improve the performance of real-time extraction of micro-motion feature in this paper based on the CPU-GPU heterogeneous parallel structure. Firstly, we discuss the principle of the micro-Doppler effect generated by the rolling of the scattering points on the orbiting satellite, analyses how to use Kalman filter to compensate the translational motion of tumbling satellite and how to use the joint time-frequency analysis and inverse Radon transform to extract the micro-motion features from the echo after compensation. Secondly, the advantages of GPU in terms of real-time processing and the working principle of CPU-GPU heterogeneous parallelism are analysed, and a program flow based on GPU to extract the micro-motion feature from the radar echo signal of rolling satellite is designed. At the end of the article the results of extraction are given to verify the correctness of the program and algorithm.

Production Level CFD Code Acceleration for Hybrid Many-Core Architectures

NASA Technical Reports Server (NTRS)

Duffy, Austen C.; Hammond, Dana P.; Nielsen, Eric J.

2012-01-01

In this work, a novel graphics processing unit (GPU) distributed sharing model for hybrid many-core architectures is introduced and employed in the acceleration of a production-level computational fluid dynamics (CFD) code. The latest generation graphics hardware allows multiple processor cores to simultaneously share a single GPU through concurrent kernel execution. This feature has allowed the NASA FUN3D code to be accelerated in parallel with up to four processor cores sharing a single GPU. For codes to scale and fully use resources on these and the next generation machines, codes will need to employ some type of GPU sharing model, as presented in this work. Findings include the effects of GPU sharing on overall performance. A discussion of the inherent challenges that parallel unstructured CFD codes face in accelerator-based computing environments is included, with considerations for future generation architectures. This work was completed by the author in August 2010, and reflects the analysis and results of the time.

Toward GPGPU accelerated human electromechanical cardiac simulations

PubMed Central

Vigueras, Guillermo; Roy, Ishani; Cookson, Andrew; Lee, Jack; Smith, Nicolas; Nordsletten, David

2014-01-01

In this paper, we look at the acceleration of weakly coupled electromechanics using the graphics processing unit (GPU). Specifically, we port to the GPU a number of components of Heart—a CPU-based finite element code developed for simulating multi-physics problems. On the basis of a criterion of computational cost, we implemented on the GPU the ODE and PDE solution steps for the electrophysiology problem and the Jacobian and residual evaluation for the mechanics problem. Performance of the GPU implementation is then compared with single core CPU (SC) execution as well as multi-core CPU (MC) computations with equivalent theoretical performance. Results show that for a human scale left ventricle mesh, GPU acceleration of the electrophysiology problem provided speedups of 164 × compared with SC and 5.5 times compared with MC for the solution of the ODE model. Speedup of up to 72 × compared with SC and 2.6 × compared with MC was also observed for the PDE solve. Using the same human geometry, the GPU implementation of mechanics residual/Jacobian computation provided speedups of up to 44 × compared with SC and 2.0 × compared with MC. © 2013 The Authors. International Journal for Numerical Methods in Biomedical Engineering published by John Wiley & Sons, Ltd. PMID:24115492

Accelerating EPI distortion correction by utilizing a modern GPU-based parallel computation.

PubMed

Yang, Yao-Hao; Huang, Teng-Yi; Wang, Fu-Nien; Chuang, Tzu-Chao; Chen, Nan-Kuei

2013-04-01

The combination of phase demodulation and field mapping is a practical method to correct echo planar imaging (EPI) geometric distortion. However, since phase dispersion accumulates in each phase-encoding step, the calculation complexity of phase modulation is Ny-fold higher than conventional image reconstructions. Thus, correcting EPI images via phase demodulation is generally a time-consuming task. Parallel computing by employing general-purpose calculations on graphics processing units (GPU) can accelerate scientific computing if the algorithm is parallelized. This study proposes a method that incorporates the GPU-based technique into phase demodulation calculations to reduce computation time. The proposed parallel algorithm was applied to a PROPELLER-EPI diffusion tensor data set. The GPU-based phase demodulation method reduced the EPI distortion correctly, and accelerated the computation. The total reconstruction time of the 16-slice PROPELLER-EPI diffusion tensor images with matrix size of 128 × 128 was reduced from 1,754 seconds to 101 seconds by utilizing the parallelized 4-GPU program. GPU computing is a promising method to accelerate EPI geometric correction. The resulting reduction in computation time of phase demodulation should accelerate postprocessing for studies performed with EPI, and should effectuate the PROPELLER-EPI technique for clinical practice. Copyright © 2011 by the American Society of Neuroimaging.

FastGCN: A GPU Accelerated Tool for Fast Gene Co-Expression Networks

PubMed Central

Liang, Meimei; Zhang, Futao; Jin, Gulei; Zhu, Jun

2015-01-01

Gene co-expression networks comprise one type of valuable biological networks. Many methods and tools have been published to construct gene co-expression networks; however, most of these tools and methods are inconvenient and time consuming for large datasets. We have developed a user-friendly, accelerated and optimized tool for constructing gene co-expression networks that can fully harness the parallel nature of GPU (Graphic Processing Unit) architectures. Genetic entropies were exploited to filter out genes with no or small expression changes in the raw data preprocessing step. Pearson correlation coefficients were then calculated. After that, we normalized these coefficients and employed the False Discovery Rate to control the multiple tests. At last, modules identification was conducted to construct the co-expression networks. All of these calculations were implemented on a GPU. We also compressed the coefficient matrix to save space. We compared the performance of the GPU implementation with those of multi-core CPU implementations with 16 CPU threads, single-thread C/C++ implementation and single-thread R implementation. Our results show that GPU implementation largely outperforms single-thread C/C++ implementation and single-thread R implementation, and GPU implementation outperforms multi-core CPU implementation when the number of genes increases. With the test dataset containing 16,000 genes and 590 individuals, we can achieve greater than 63 times the speed using a GPU implementation compared with a single-thread R implementation when 50 percent of genes were filtered out and about 80 times the speed when no genes were filtered out. PMID:25602758

FastGCN: a GPU accelerated tool for fast gene co-expression networks.

PubMed

Liang, Meimei; Zhang, Futao; Jin, Gulei; Zhu, Jun

2015-01-01

Gene co-expression networks comprise one type of valuable biological networks. Many methods and tools have been published to construct gene co-expression networks; however, most of these tools and methods are inconvenient and time consuming for large datasets. We have developed a user-friendly, accelerated and optimized tool for constructing gene co-expression networks that can fully harness the parallel nature of GPU (Graphic Processing Unit) architectures. Genetic entropies were exploited to filter out genes with no or small expression changes in the raw data preprocessing step. Pearson correlation coefficients were then calculated. After that, we normalized these coefficients and employed the False Discovery Rate to control the multiple tests. At last, modules identification was conducted to construct the co-expression networks. All of these calculations were implemented on a GPU. We also compressed the coefficient matrix to save space. We compared the performance of the GPU implementation with those of multi-core CPU implementations with 16 CPU threads, single-thread C/C++ implementation and single-thread R implementation. Our results show that GPU implementation largely outperforms single-thread C/C++ implementation and single-thread R implementation, and GPU implementation outperforms multi-core CPU implementation when the number of genes increases. With the test dataset containing 16,000 genes and 590 individuals, we can achieve greater than 63 times the speed using a GPU implementation compared with a single-thread R implementation when 50 percent of genes were filtered out and about 80 times the speed when no genes were filtered out.

Advantages of GPU technology in DFT calculations of intercalated graphene

NASA Astrophysics Data System (ADS)

Pešić, J.; Gajić, R.

2014-09-01

Over the past few years, the expansion of general-purpose graphic-processing unit (GPGPU) technology has had a great impact on computational science. GPGPU is the utilization of a graphics-processing unit (GPU) to perform calculations in applications usually handled by the central processing unit (CPU). Use of GPGPUs as a way to increase computational power in the material sciences has significantly decreased computational costs in already highly demanding calculations. A level of the acceleration and parallelization depends on the problem itself. Some problems can benefit from GPU acceleration and parallelization, such as the finite-difference time-domain algorithm (FTDT) and density-functional theory (DFT), while others cannot take advantage of these modern technologies. A number of GPU-supported applications had emerged in the past several years (www.nvidia.com/object/gpu-applications.html). Quantum Espresso (QE) is reported as an integrated suite of open source computer codes for electronic-structure calculations and materials modeling at the nano-scale. It is based on DFT, the use of a plane-waves basis and a pseudopotential approach. Since the QE 5.0 version, it has been implemented as a plug-in component for standard QE packages that allows exploiting the capabilities of Nvidia GPU graphic cards (www.qe-forge.org/gf/proj). In this study, we have examined the impact of the usage of GPU acceleration and parallelization on the numerical performance of DFT calculations. Graphene has been attracting attention worldwide and has already shown some remarkable properties. We have studied an intercalated graphene, using the QE package PHonon, which employs GPU. The term ‘intercalation’ refers to a process whereby foreign adatoms are inserted onto a graphene lattice. In addition, by intercalating different atoms between graphene layers, it is possible to tune their physical properties. Our experiments have shown there are benefits from using GPUs, and we reached an acceleration of several times compared to standard CPU calculations.

Evaluating the Usefulness of a Novel 10B-Carrier Conjugated With Cyclic RGD Peptide in Boron Neutron Capture Therapy

PubMed Central

Masunaga, Shin-ichiro; Kimura, Sadaaki; Harada, Tomohiro; Okuda, Kensuke; Sakurai, Yoshinori; Tanaka, Hiroki; Suzuki, Minoru; Kondo, Natsuko; Maruhashi, Akira; Nagasawa, Hideko; Ono, Koji

2012-01-01

Background To evaluate the usefulness of a novel 10B-carrier conjugated with an integrin-binding cyclic RGD peptide (GPU-201) in boron neutron capture therapy (BNCT). Methods GPU-201 was synthesized from integrin-binding Arg-Gly-Asp (RGD) consensus sequence of matrix proteins and a 10B cluster 1, 2-dicarba-closo-dodecaborane-10B. Mercaptododecaborate-10B (BSH) dissolved in physiological saline and BSH and GPU-201 dissolved with cyclodextrin (CD) as a solubilizing and dispersing agent were intraperitoneally administered to SCC VII tumor-bearing mice. Then, the 10B concentrations in the tumors and normal tissues were measured by γ-ray spectrometry. Meanwhile, tumor-bearing mice were continuously given 5-bromo-2’-deoxyuridine (BrdU) to label all proliferating (P) cells in the tumors, then treated with GPU-201, BSH-CD, or BSH. Immediately after reactor neutron beam or γ-ray irradiation, during which intratumor 10B concentrations were kept at levels similar to each other, cells from some tumors were isolated and incubated with a cytokinesis blocker. The responses of the Q and total (= P + Q) cell populations were assessed based on the frequency of micronuclei using immunofluorescence staining for BrdU. Results The 10B from BSH was washed away rapidly in all these tissues and the retention of 10B from BSH-CD and GPU-201 was similar except in blood where the 10B concentration from GPU-201 was higher for longer. GPU-201 showed a significantly stronger radio-sensitizing effect under neutron beam irradiation on both total and Q cell populations than any other 10B-carrier. Conclusion A novel 10B-carrier conjugated with an integrin-binding RGD peptide (GPU-201) that sensitized tumor cells more markedly than conventional 10B-carriers may be a promising candidate for use in BNCT. However, its toxicity needs to be tested further. PMID:29147290

High performance MRI simulations of motion on multi-GPU systems.

PubMed

Xanthis, Christos G; Venetis, Ioannis E; Aletras, Anthony H

2014-07-04

MRI physics simulators have been developed in the past for optimizing imaging protocols and for training purposes. However, these simulators have only addressed motion within a limited scope. The purpose of this study was the incorporation of realistic motion, such as cardiac motion, respiratory motion and flow, within MRI simulations in a high performance multi-GPU environment. Three different motion models were introduced in the Magnetic Resonance Imaging SIMULator (MRISIMUL) of this study: cardiac motion, respiratory motion and flow. Simulation of a simple Gradient Echo pulse sequence and a CINE pulse sequence on the corresponding anatomical model was performed. Myocardial tagging was also investigated. In pulse sequence design, software crushers were introduced to accommodate the long execution times in order to avoid spurious echoes formation. The displacement of the anatomical model isochromats was calculated within the Graphics Processing Unit (GPU) kernel for every timestep of the pulse sequence. Experiments that would allow simulation of custom anatomical and motion models were also performed. Last, simulations of motion with MRISIMUL on single-node and multi-node multi-GPU systems were examined. Gradient Echo and CINE images of the three motion models were produced and motion-related artifacts were demonstrated. The temporal evolution of the contractility of the heart was presented through the application of myocardial tagging. Better simulation performance and image quality were presented through the introduction of software crushers without the need to further increase the computational load and GPU resources. Last, MRISIMUL demonstrated an almost linear scalable performance with the increasing number of available GPU cards, in both single-node and multi-node multi-GPU computer systems. MRISIMUL is the first MR physics simulator to have implemented motion with a 3D large computational load on a single computer multi-GPU configuration. The incorporation of realistic motion models, such as cardiac motion, respiratory motion and flow may benefit the design and optimization of existing or new MR pulse sequences, protocols and algorithms, which examine motion related MR applications.

A Framework for Batched and GPU-Resident Factorization Algorithms Applied to Block Householder Transformations

DOE Office of Scientific and Technical Information (OSTI.GOV)

Dong, Tingzing Tim; Tomov, Stanimire Z; Luszczek, Piotr R

As modern hardware keeps evolving, an increasingly effective approach to developing energy efficient and high-performance solvers is to design them to work on many small size and independent problems. Many applications already need this functionality, especially for GPUs, which are currently known to be about four to five times more energy efficient than multicore CPUs. We describe the development of one-sided factorizations that work for a set of small dense matrices in parallel, and we illustrate our techniques on the QR factorization based on Householder transformations. We refer to this mode of operation as a batched factorization. Our approach ismore » based on representing the algorithms as a sequence of batched BLAS routines for GPU-only execution. This is in contrast to the hybrid CPU-GPU algorithms that rely heavily on using the multicore CPU for specific parts of the workload. But for a system to benefit fully from the GPU's significantly higher energy efficiency, avoiding the use of the multicore CPU must be a primary design goal, so the system can rely more heavily on the more efficient GPU. Additionally, this will result in the removal of the costly CPU-to-GPU communication. Furthermore, we do not use a single symmetric multiprocessor(on the GPU) to factorize a single problem at a time. We illustrate how our performance analysis, and the use of profiling and tracing tools, guided the development and optimization of our batched factorization to achieve up to a 2-fold speedup and a 3-fold energy efficiency improvement compared to our highly optimized batched CPU implementations based on the MKL library(when using two sockets of Intel Sandy Bridge CPUs). Compared to a batched QR factorization featured in the CUBLAS library for GPUs, we achieved up to 5x speedup on the K40 GPU.« less

Democratic Population Decisions Result in Robust Policy-Gradient Learning: A Parametric Study with GPU Simulations

PubMed Central

Richmond, Paul; Buesing, Lars; Giugliano, Michele; Vasilaki, Eleni

2011-01-01

High performance computing on the Graphics Processing Unit (GPU) is an emerging field driven by the promise of high computational power at a low cost. However, GPU programming is a non-trivial task and moreover architectural limitations raise the question of whether investing effort in this direction may be worthwhile. In this work, we use GPU programming to simulate a two-layer network of Integrate-and-Fire neurons with varying degrees of recurrent connectivity and investigate its ability to learn a simplified navigation task using a policy-gradient learning rule stemming from Reinforcement Learning. The purpose of this paper is twofold. First, we want to support the use of GPUs in the field of Computational Neuroscience. Second, using GPU computing power, we investigate the conditions under which the said architecture and learning rule demonstrate best performance. Our work indicates that networks featuring strong Mexican-Hat-shaped recurrent connections in the top layer, where decision making is governed by the formation of a stable activity bump in the neural population (a “non-democratic” mechanism), achieve mediocre learning results at best. In absence of recurrent connections, where all neurons “vote” independently (“democratic”) for a decision via population vector readout, the task is generally learned better and more robustly. Our study would have been extremely difficult on a desktop computer without the use of GPU programming. We present the routines developed for this purpose and show that a speed improvement of 5x up to 42x is provided versus optimised Python code. The higher speed is achieved when we exploit the parallelism of the GPU in the search of learning parameters. This suggests that efficient GPU programming can significantly reduce the time needed for simulating networks of spiking neurons, particularly when multiple parameter configurations are investigated. PMID:21572529

GPU accelerated generation of digitally reconstructed radiographs for 2-D/3-D image registration.

PubMed

Dorgham, Osama M; Laycock, Stephen D; Fisher, Mark H

2012-09-01

Recent advances in programming languages for graphics processing units (GPUs) provide developers with a convenient way of implementing applications which can be executed on the CPU and GPU interchangeably. GPUs are becoming relatively cheap, powerful, and widely available hardware components, which can be used to perform intensive calculations. The last decade of hardware performance developments shows that GPU-based computation is progressing significantly faster than CPU-based computation, particularly if one considers the execution of highly parallelisable algorithms. Future predictions illustrate that this trend is likely to continue. In this paper, we introduce a way of accelerating 2-D/3-D image registration by developing a hybrid system which executes on the CPU and utilizes the GPU for parallelizing the generation of digitally reconstructed radiographs (DRRs). Based on the advancements of the GPU over the CPU, it is timely to exploit the benefits of many-core GPU technology by developing algorithms for DRR generation. Although some previous work has investigated the rendering of DRRs using the GPU, this paper investigates approximations which reduce the computational overhead while still maintaining a quality consistent with that needed for 2-D/3-D registration with sufficient accuracy to be clinically acceptable in certain applications of radiation oncology. Furthermore, by comparing implementations of 2-D/3-D registration on the CPU and GPU, we investigate current performance and propose an optimal framework for PC implementations addressing the rigid registration problem. Using this framework, we are able to render DRR images from a 256×256×133 CT volume in ~24 ms using an NVidia GeForce 8800 GTX and in ~2 ms using NVidia GeForce GTX 580. In addition to applications requiring fast automatic patient setup, these levels of performance suggest image-guided radiation therapy at video frame rates is technically feasible using relatively low cost PC architecture.

Roofline model toolkit: A practical tool for architectural and program analysis

DOE Office of Scientific and Technical Information (OSTI.GOV)

Lo, Yu Jung; Williams, Samuel; Van Straalen, Brian

We present preliminary results of the Roofline Toolkit for multicore, many core, and accelerated architectures. This paper focuses on the processor architecture characterization engine, a collection of portable instrumented micro benchmarks implemented with Message Passing Interface (MPI), and OpenMP used to express thread-level parallelism. These benchmarks are specialized to quantify the behavior of different architectural features. Compared to previous work on performance characterization, these microbenchmarks focus on capturing the performance of each level of the memory hierarchy, along with thread-level parallelism, instruction-level parallelism and explicit SIMD parallelism, measured in the context of the compilers and run-time environments. We also measuremore » sustained PCIe throughput with four GPU memory managed mechanisms. By combining results from the architecture characterization with the Roofline model based solely on architectural specifications, this work offers insights for performance prediction of current and future architectures and their software systems. To that end, we instrument three applications and plot their resultant performance on the corresponding Roofline model when run on a Blue Gene/Q architecture.« less

Dataflow-Based Implementation of Layered Sensing Applications on High-Performance Embedded Processors

DTIC Science & Technology

2013-03-01

time (milliseconds) GFlops Comparison to GPU peak performance (%) Cascade Gaussian Filtering 13 45.19 6.3 Difference of Gaussian 0.512 152...values for the GPU-targeted actor implementations in terms of Giga Floating Point Operations Per Second ( GFLOPS ). Our GFLOPS calculation for an actor...kernels. The results for GFLOPS are provided in Table . The actors were implemented on an NVIDIA GTX260 GPU, which provides 715 GFLOPS as peak

GPU based framework for geospatial analyses

NASA Astrophysics Data System (ADS)

Cosmin Sandric, Ionut; Ionita, Cristian; Dardala, Marian; Furtuna, Titus

2017-04-01

Parallel processing on multiple CPU cores is already used at large scale in geocomputing, but parallel processing on graphics cards is just at the beginning. Being able to use an simple laptop with a dedicated graphics card for advanced and very fast geocomputation is an advantage that each scientist wants to have. The necessity to have high speed computation in geosciences has increased in the last 10 years, mostly due to the increase in the available datasets. These datasets are becoming more and more detailed and hence they require more space to store and more time to process. Distributed computation on multicore CPU's and GPU's plays an important role by processing one by one small parts from these big datasets. These way of computations allows to speed up the process, because instead of using just one process for each dataset, the user can use all the cores from a CPU or up to hundreds of cores from GPU The framework provide to the end user a standalone tools for morphometry analyses at multiscale level. An important part of the framework is dedicated to uncertainty propagation in geospatial analyses. The uncertainty may come from the data collection or may be induced by the model or may have an infinite sources. These uncertainties plays important roles when a spatial delineation of the phenomena is modelled. Uncertainty propagation is implemented inside the GPU framework using Monte Carlo simulations. The GPU framework with the standalone tools proved to be a reliable tool for modelling complex natural phenomena The framework is based on NVidia Cuda technology and is written in C++ programming language. The code source will be available on github at https://github.com/sandricionut/GeoRsGPU Acknowledgement: GPU framework for geospatial analysis, Young Researchers Grant (ICUB-University of Bucharest) 2016, director Ionut Sandric

Implementation and optimization of ultrasound signal processing algorithms on mobile GPU

NASA Astrophysics Data System (ADS)

Kong, Woo Kyu; Lee, Wooyoul; Kim, Kyu Cheol; Yoo, Yangmo; Song, Tai-Kyong

2014-03-01

A general-purpose graphics processing unit (GPGPU) has been used for improving computing power in medical ultrasound imaging systems. Recently, a mobile GPU becomes powerful to deal with 3D games and videos at high frame rates on Full HD or HD resolution displays. This paper proposes the method to implement ultrasound signal processing on a mobile GPU available in the high-end smartphone (Galaxy S4, Samsung Electronics, Seoul, Korea) with programmable shaders on the OpenGL ES 2.0 platform. To maximize the performance of the mobile GPU, the optimization of shader design and load sharing between vertex and fragment shader was performed. The beamformed data were captured from a tissue mimicking phantom (Model 539 Multipurpose Phantom, ATS Laboratories, Inc., Bridgeport, CT, USA) by using a commercial ultrasound imaging system equipped with a research package (Ultrasonix Touch, Ultrasonix, Richmond, BC, Canada). The real-time performance is evaluated by frame rates while varying the range of signal processing blocks. The implementation method of ultrasound signal processing on OpenGL ES 2.0 was verified by analyzing PSNR with MATLAB gold standard that has the same signal path. CNR was also analyzed to verify the method. From the evaluations, the proposed mobile GPU-based processing method has no significant difference with the processing using MATLAB (i.e., PSNR<52.51 dB). The comparable results of CNR were obtained from both processing methods (i.e., 11.31). From the mobile GPU implementation, the frame rates of 57.6 Hz were achieved. The total execution time was 17.4 ms that was faster than the acquisition time (i.e., 34.4 ms). These results indicate that the mobile GPU-based processing method can support real-time ultrasound B-mode processing on the smartphone.

GPU-Accelerated Voxelwise Hepatic Perfusion Quantification

PubMed Central

Wang, H; Cao, Y

2012-01-01

Voxelwise quantification of hepatic perfusion parameters from dynamic contrast enhanced (DCE) imaging greatly contributes to assessment of liver function in response to radiation therapy. However, the efficiency of the estimation of hepatic perfusion parameters voxel-by-voxel in the whole liver using a dual-input single-compartment model requires substantial improvement for routine clinical applications. In this paper, we utilize the parallel computation power of a graphics processing unit (GPU) to accelerate the computation, while maintaining the same accuracy as the conventional method. Using CUDA-GPU, the hepatic perfusion computations over multiple voxels are run across the GPU blocks concurrently but independently. At each voxel, non-linear least squares fitting the time series of the liver DCE data to the compartmental model is distributed to multiple threads in a block, and the computations of different time points are performed simultaneously and synchronically. An efficient fast Fourier transform in a block is also developed for the convolution computation in the model. The GPU computations of the voxel-by-voxel hepatic perfusion images are compared with ones by the CPU using the simulated DCE data and the experimental DCE MR images from patients. The computation speed is improved by 30 times using a NVIDIA Tesla C2050 GPU compared to a 2.67 GHz Intel Xeon CPU processor. To obtain liver perfusion maps with 626400 voxels in a patient’s liver, it takes 0.9 min with the GPU-accelerated voxelwise computation, compared to 110 min with the CPU, while both methods result in perfusion parameters differences less than 10−6. The method will be useful for generating liver perfusion images in clinical settings. PMID:22892645

Fast-GPU-PCC: A GPU-Based Technique to Compute Pairwise Pearson's Correlation Coefficients for Time Series Data-fMRI Study.

PubMed

Eslami, Taban; Saeed, Fahad

2018-04-20

Functional magnetic resonance imaging (fMRI) is a non-invasive brain imaging technique, which has been regularly used for studying brain’s functional activities in the past few years. A very well-used measure for capturing functional associations in brain is Pearson’s correlation coefficient. Pearson’s correlation is widely used for constructing functional network and studying dynamic functional connectivity of the brain. These are useful measures for understanding the effects of brain disorders on connectivities among brain regions. The fMRI scanners produce huge number of voxels and using traditional central processing unit (CPU)-based techniques for computing pairwise correlations is very time consuming especially when large number of subjects are being studied. In this paper, we propose a graphics processing unit (GPU)-based algorithm called Fast-GPU-PCC for computing pairwise Pearson’s correlation coefficient. Based on the symmetric property of Pearson’s correlation, this approach returns N ( N − 1 ) / 2 correlation coefficients located at strictly upper triangle part of the correlation matrix. Storing correlations in a one-dimensional array with the order as proposed in this paper is useful for further usage. Our experiments on real and synthetic fMRI data for different number of voxels and varying length of time series show that the proposed approach outperformed state of the art GPU-based techniques as well as the sequential CPU-based versions. We show that Fast-GPU-PCC runs 62 times faster than CPU-based version and about 2 to 3 times faster than two other state of the art GPU-based methods.

SU-D-206-01: Employing a Novel Consensus Optimization Strategy to Achieve Iterative Cone Beam CT Reconstruction On a Multi-GPU Platform

DOE Office of Scientific and Technical Information (OSTI.GOV)

Li, B; Southern Medical University, Guangzhou, Guangdong; Tian, Z

Purpose: While compressed sensing-based cone-beam CT (CBCT) iterative reconstruction techniques have demonstrated tremendous capability of reconstructing high-quality images from undersampled noisy data, its long computation time still hinders wide application in routine clinic. The purpose of this study is to develop a reconstruction framework that employs modern consensus optimization techniques to achieve CBCT reconstruction on a multi-GPU platform for improved computational efficiency. Methods: Total projection data were evenly distributed to multiple GPUs. Each GPU performed reconstruction using its own projection data with a conventional total variation regularization approach to ensure image quality. In addition, the solutions from GPUs were subjectmore » to a consistency constraint that they should be identical. We solved the optimization problem with all the constraints considered rigorously using an alternating direction method of multipliers (ADMM) algorithm. The reconstruction framework was implemented using OpenCL on a platform with two Nvidia GTX590 GPU cards, each with two GPUs. We studied the performance of our method and demonstrated its advantages through a simulation case with a NCAT phantom and an experimental case with a Catphan phantom. Result: Compared with the CBCT images reconstructed using conventional FDK method with full projection datasets, our proposed method achieved comparable image quality with about one third projection numbers. The computation time on the multi-GPU platform was ∼55 s and ∼ 35 s in the two cases respectively, achieving a speedup factor of ∼ 3.0 compared with single GPU reconstruction. Conclusion: We have developed a consensus ADMM-based CBCT reconstruction method which enabled performing reconstruction on a multi-GPU platform. The achieved efficiency made this method clinically attractive.« less

Efficient Parallel Levenberg-Marquardt Model Fitting towards Real-Time Automated Parametric Imaging Microscopy

PubMed Central

Zhu, Xiang; Zhang, Dianwen

2013-01-01

We present a fast, accurate and robust parallel Levenberg-Marquardt minimization optimizer, GPU-LMFit, which is implemented on graphics processing unit for high performance scalable parallel model fitting processing. GPU-LMFit can provide a dramatic speed-up in massive model fitting analyses to enable real-time automated pixel-wise parametric imaging microscopy. We demonstrate the performance of GPU-LMFit for the applications in superresolution localization microscopy and fluorescence lifetime imaging microscopy. PMID:24130785

Problems Related to Parallelization of CFD Algorithms on GPU, Multi-GPU and Hybrid Architectures

NASA Astrophysics Data System (ADS)

Biazewicz, Marek; Kurowski, Krzysztof; Ludwiczak, Bogdan; Napieraia, Krystyna

2010-09-01

Computational Fluid Dynamics (CFD) is one of the branches of fluid mechanics, which uses numerical methods and algorithms to solve and analyze fluid flows. CFD is used in various domains, such as oil and gas reservoir uncertainty analysis, aerodynamic body shapes optimization (e.g. planes, cars, ships, sport helmets, skis), natural phenomena analysis, numerical simulation for weather forecasting or realistic visualizations. CFD problem is very complex and needs a lot of computational power to obtain the results in a reasonable time. We have implemented a parallel application for two-dimensional CFD simulation with a free surface approximation (MAC method) using new hardware architectures, in particular multi-GPU and hybrid computing environments. For this purpose we decided to use NVIDIA graphic cards with CUDA environment due to its simplicity of programming and good computations performance. We used finite difference discretization of Navier-Stokes equations, where fluid is propagated over an Eulerian Grid. In this model, the behavior of the fluid inside the cell depends only on the properties of local, surrounding cells, therefore it is well suited for the GPU-based architecture. In this paper we demonstrate how to use efficiently the computing power of GPUs for CFD. Additionally, we present some best practices to help users analyze and improve the performance of CFD applications executed on GPU. Finally, we discuss various challenges around the multi-GPU implementation on the example of matrix multiplication.

Improving metabolic stability with deuterium: The discovery of GPU-028, a potent free fatty acid receptor 4 agonists.

PubMed

Li, Zheng; Xu, Xue; Li, Gang; Fu, Xiaoting; Liu, Yanzhi; Feng, Yufeng; Wang, Mingyan; Ouyang, Yunting; Han, Jing

2017-12-15

The free fatty acid receptor 4 (FFA4) has emerged as a promising anti-diabetic target due to its function in improvement of insulin secretion and insulin resistance. The FFA4 agonist TUG-891 revealed great potential as a widely used pharmacological tool, but it has been suffered from high plasma clearance probably because the phenylpropanoic acid is vulnerable to β-oxidation. To identify metabolically stable analog without influence on physiological mechanism of TUG-891, we tried to incorporate deuterium at the α-position of phenylpropionic acid to afford compound 4 (GPU-028). As expected, GPU-028 revealed a longer half-life (T 1/2  = 1.66 h), lower clearance (CL = 0.97 L/h/kg) and higher maximum plasma concentration (C max  = 2035.23 μg/L), resulting in a 4-fold higher exposure than TUG-891. Although GPU-028 exhibited a similar agonistic activity in comparison to TUG-891, the hypoglycemic effect of GPU-028 was better than that of TUG-891 after treatment over four weeks in diet-induced obese mice. These positive results indicated that GPU-028 might be a better pharmacological tool than TUG-891 to explore physiological function of FFA4, especially on the in vivo study. Copyright © 2017 Elsevier Ltd. All rights reserved.

Higher-order ice-sheet modelling accelerated by multigrid on graphics cards

NASA Astrophysics Data System (ADS)

Brædstrup, Christian; Egholm, David

2013-04-01

Higher-order ice flow modelling is a very computer intensive process owing primarily to the nonlinear influence of the horizontal stress coupling. When applied for simulating long-term glacial landscape evolution, the ice-sheet models must consider very long time series, while both high temporal and spatial resolution is needed to resolve small effects. The use of higher-order and full stokes models have therefore seen very limited usage in this field. However, recent advances in graphics card (GPU) technology for high performance computing have proven extremely efficient in accelerating many large-scale scientific computations. The general purpose GPU (GPGPU) technology is cheap, has a low power consumption and fits into a normal desktop computer. It could therefore provide a powerful tool for many glaciologists working on ice flow models. Our current research focuses on utilising the GPU as a tool in ice-sheet and glacier modelling. To this extent we have implemented the Integrated Second-Order Shallow Ice Approximation (iSOSIA) equations on the device using the finite difference method. To accelerate the computations, the GPU solver uses a non-linear Red-Black Gauss-Seidel iterator coupled with a Full Approximation Scheme (FAS) multigrid setup to further aid convergence. The GPU finite difference implementation provides the inherent parallelization that scales from hundreds to several thousands of cores on newer cards. We demonstrate the efficiency of the GPU multigrid solver using benchmark experiments.

The novel implicit LU-SGS parallel iterative method based on the diffusion equation of a nuclear reactor on a GPU cluster

NASA Astrophysics Data System (ADS)

Zhang, Jilin; Sha, Chaoqun; Wu, Yusen; Wan, Jian; Zhou, Li; Ren, Yongjian; Si, Huayou; Yin, Yuyu; Jing, Ya

2017-02-01

GPU not only is used in the field of graphic technology but also has been widely used in areas needing a large number of numerical calculations. In the energy industry, because of low carbon, high energy density, high duration and other characteristics, the development of nuclear energy cannot easily be replaced by other energy sources. Management of core fuel is one of the major areas of concern in a nuclear power plant, and it is directly related to the economic benefits and cost of nuclear power. The large-scale reactor core expansion equation is large and complicated, so the calculation of the diffusion equation is crucial in the core fuel management process. In this paper, we use CUDA programming technology on a GPU cluster to run the LU-SGS parallel iterative calculation against the background of the diffusion equation of the reactor. We divide one-dimensional and two-dimensional mesh into a plurality of domains, with each domain evenly distributed on the GPU blocks. A parallel collision scheme is put forward that defines the virtual boundary of the grid exchange information and data transmission by non-stop collision. Compared with the serial program, the experiment shows that GPU greatly improves the efficiency of program execution and verifies that GPU is playing a much more important role in the field of numerical calculations.

permGPU: Using graphics processing units in RNA microarray association studies.

PubMed

Shterev, Ivo D; Jung, Sin-Ho; George, Stephen L; Owzar, Kouros

2010-06-16

Many analyses of microarray association studies involve permutation, bootstrap resampling and cross-validation, that are ideally formulated as embarrassingly parallel computing problems. Given that these analyses are computationally intensive, scalable approaches that can take advantage of multi-core processor systems need to be developed. We have developed a CUDA based implementation, permGPU, that employs graphics processing units in microarray association studies. We illustrate the performance and applicability of permGPU within the context of permutation resampling for a number of test statistics. An extensive simulation study demonstrates a dramatic increase in performance when using permGPU on an NVIDIA GTX 280 card compared to an optimized C/C++ solution running on a conventional Linux server. permGPU is available as an open-source stand-alone application and as an extension package for the R statistical environment. It provides a dramatic increase in performance for permutation resampling analysis in the context of microarray association studies. The current version offers six test statistics for carrying out permutation resampling analyses for binary, quantitative and censored time-to-event traits.

Parallel Computer System for 3D Visualization Stereo on GPU

NASA Astrophysics Data System (ADS)

Al-Oraiqat, Anas M.; Zori, Sergii A.

2018-03-01

This paper proposes the organization of a parallel computer system based on Graphic Processors Unit (GPU) for 3D stereo image synthesis. The development is based on the modified ray tracing method developed by the authors for fast search of tracing rays intersections with scene objects. The system allows significant increase in the productivity for the 3D stereo synthesis of photorealistic quality. The generalized procedure of 3D stereo image synthesis on the Graphics Processing Unit/Graphics Processing Clusters (GPU/GPC) is proposed. The efficiency of the proposed solutions by GPU implementation is compared with single-threaded and multithreaded implementations on the CPU. The achieved average acceleration in multi-thread implementation on the test GPU and CPU is about 7.5 and 1.6 times, respectively. Studying the influence of choosing the size and configuration of the computational Compute Unified Device Archi-tecture (CUDA) network on the computational speed shows the importance of their correct selection. The obtained experimental estimations can be significantly improved by new GPUs with a large number of processing cores and multiprocessors, as well as optimized configuration of the computing CUDA network.

Multi-GPU parallel algorithm design and analysis for improved inversion of probability tomography with gravity gradiometry data

NASA Astrophysics Data System (ADS)

Hou, Zhenlong; Huang, Danian

2017-09-01

In this paper, we make a study on the inversion of probability tomography (IPT) with gravity gradiometry data at first. The space resolution of the results is improved by multi-tensor joint inversion, depth weighting matrix and the other methods. Aiming at solving the problems brought by the big data in the exploration, we present the parallel algorithm and the performance analysis combining Compute Unified Device Architecture (CUDA) with Open Multi-Processing (OpenMP) based on Graphics Processing Unit (GPU) accelerating. In the test of the synthetic model and real data from Vinton Dome, we get the improved results. It is also proved that the improved inversion algorithm is effective and feasible. The performance of parallel algorithm we designed is better than the other ones with CUDA. The maximum speedup could be more than 200. In the performance analysis, multi-GPU speedup and multi-GPU efficiency are applied to analyze the scalability of the multi-GPU programs. The designed parallel algorithm is demonstrated to be able to process larger scale of data and the new analysis method is practical.

Implementation of EAM and FS potentials in HOOMD-blue

NASA Astrophysics Data System (ADS)

Yang, Lin; Zhang, Feng; Travesset, Alex; Wang, Caizhuang; Ho, Kaiming

HOOMD-blue is a general-purpose software to perform classical molecular dynamics simulations entirely on GPUs. We provide full support for EAM and FS type potentials in HOOMD-blue, and report accuracy and efficiency benchmarks, including comparisons with the LAMMPS GPU package. Two problems were selected to test the accuracy: the determination of the glass transition temperature of Cu64.5Zr35.5 alloy using an FS potential and the calculation of pair distribution functions of Ni3Al using an EAM potential. In both cases, the results using HOOMD-blue are indistinguishable from those obtained by the GPU package in LAMMPS within statistical uncertainties. As tests for time efficiency, we benchmark time-steps per second using LAMMPS GPU and HOOMD-blue on one NVIDIA Tesla GPU. Compared to our typical LAMMPS simulations on one CPU cluster node which has 16 CPUs, LAMMPS GPU can be 3-3.5 times faster, and HOOMD-blue can be 4-5.5 times faster. We acknowledge the support from Laboratory Directed Research and Development (LDRD) of Ames Laboratory.

GPU-accelerated FDTD modeling of radio-frequency field-tissue interactions in high-field MRI.

PubMed

Chi, Jieru; Liu, Feng; Weber, Ewald; Li, Yu; Crozier, Stuart

2011-06-01

The analysis of high-field RF field-tissue interactions requires high-performance finite-difference time-domain (FDTD) computing. Conventional CPU-based FDTD calculations offer limited computing performance in a PC environment. This study presents a graphics processing unit (GPU)-based parallel-computing framework, producing substantially boosted computing efficiency (with a two-order speedup factor) at a PC-level cost. Specific details of implementing the FDTD method on a GPU architecture have been presented and the new computational strategy has been successfully applied to the design of a novel 8-element transceive RF coil system at 9.4 T. Facilitated by the powerful GPU-FDTD computing, the new RF coil array offers optimized fields (averaging 25% improvement in sensitivity, and 20% reduction in loop coupling compared with conventional array structures of the same size) for small animal imaging with a robust RF configuration. The GPU-enabled acceleration paves the way for FDTD to be applied for both detailed forward modeling and inverse design of MRI coils, which were previously impractical.

Gctf: Real-time CTF determination and correction

PubMed Central

Zhang, Kai

2016-01-01

Accurate estimation of the contrast transfer function (CTF) is critical for a near-atomic resolution cryo electron microscopy (cryoEM) reconstruction. Here, a GPU-accelerated computer program, Gctf, for accurate and robust, real-time CTF determination is presented. The main target of Gctf is to maximize the cross-correlation of a simulated CTF with the logarithmic amplitude spectra (LAS) of observed micrographs after background subtraction. Novel approaches in Gctf improve both speed and accuracy. In addition to GPU acceleration (e.g. 10–50×), a fast ‘1-dimensional search plus 2-dimensional refinement (1S2R)’ procedure further speeds up Gctf. Based on the global CTF determination, the local defocus for each particle and for single frames of movies is accurately refined, which improves CTF parameters of all particles for subsequent image processing. Novel diagnosis method using equiphase averaging (EPA) and self-consistency verification procedures have also been implemented in the program for practical use, especially for aims of near-atomic reconstruction. Gctf is an independent program and the outputs can be easily imported into other cryoEM software such as Relion (Scheres, 2012) and Frealign (Grigorieff, 2007). The results from several representative datasets are shown and discussed in this paper. PMID:26592709

Ramses-GPU: Second order MUSCL-Handcock finite volume fluid solver

NASA Astrophysics Data System (ADS)

Kestener, Pierre

2017-10-01

RamsesGPU is a reimplementation of RAMSES (ascl:1011.007) which drops the adaptive mesh refinement (AMR) features to optimize 3D uniform grid algorithms for modern graphics processor units (GPU) to provide an efficient software package for astrophysics applications that do not need AMR features but do require a very large number of integration time steps. RamsesGPU provides an very efficient C++/CUDA/MPI software implementation of a second order MUSCL-Handcock finite volume fluid solver for compressible hydrodynamics as a magnetohydrodynamics solver based on the constraint transport technique. Other useful modules includes static gravity, dissipative terms (viscosity, resistivity), and forcing source term for turbulence studies, and special care was taken to enhance parallel input/output performance by using state-of-the-art libraries such as HDF5 and parallel-netcdf.

Simulation of ring polymer melts with GPU acceleration

NASA Astrophysics Data System (ADS)

Schram, R. D.; Barkema, G. T.

2018-06-01

We implemented the elastic lattice polymer model on the GPU (Graphics Processing Unit), and show that the GPU is very efficient for polymer simulations of dense polymer melts. The implementation is able to perform up to 4.1 ṡ109 Monte Carlo moves per second. Compared to our standard CPU implementation, we find an effective speed-up of a factor 92. Using this GPU implementation we studied the equilibrium properties and the dynamics of non-concatenated ring polymers in a melt of such polymers, using Rouse modes. With increasing polymer length, we found a very slow transition to compactness with a growth exponent ν ≈ 1 / 3. Numerically we find that the longest internal time scale of the polymer scales as N3.1, with N the molecular weight of the ring polymer.

Tensor Algebra Library for NVidia Graphics Processing Units

DOE Office of Scientific and Technical Information (OSTI.GOV)

Liakh, Dmitry

This is a general purpose math library implementing basic tensor algebra operations on NVidia GPU accelerators. This software is a tensor algebra library that can perform basic tensor algebra operations, including tensor contractions, tensor products, tensor additions, etc., on NVidia GPU accelerators, asynchronously with respect to the CPU host. It supports a simultaneous use of multiple NVidia GPUs. Each asynchronous API function returns a handle which can later be used for querying the completion of the corresponding tensor algebra operation on a specific GPU. The tensors participating in a particular tensor operation are assumed to be stored in local RAMmore » of a node or GPU RAM. The main research area where this library can be utilized is the quantum many-body theory (e.g., in electronic structure theory).« less

Overview of implementation of DARPA GPU program in SAIC

NASA Astrophysics Data System (ADS)

Braunreiter, Dennis; Furtek, Jeremy; Chen, Hai-Wen; Healy, Dennis

2008-04-01

This paper reviews the implementation of DARPA MTO STAP-BOY program for both Phase I and II conducted at Science Applications International Corporation (SAIC). The STAP-BOY program conducts fast covariance factorization and tuning techniques for space-time adaptive process (STAP) Algorithm Implementation on Graphics Processor unit (GPU) Architectures for Embedded Systems. The first part of our presentation on the DARPA STAP-BOY program will focus on GPU implementation and algorithm innovations for a prototype radar STAP algorithm. The STAP algorithm will be implemented on the GPU, using stream programming (from companies such as PeakStream, ATI Technologies' CTM, and NVIDIA) and traditional graphics APIs. This algorithm will include fast range adaptive STAP weight updates and beamforming applications, each of which has been modified to exploit the parallel nature of graphics architectures.

Multi-core and GPU accelerated simulation of a radial star target imaged with equivalent t-number circular and Gaussian pupils

NASA Astrophysics Data System (ADS)

Greynolds, Alan W.

2013-09-01

Results from the GelOE optical engineering software are presented for the through-focus, monochromatic coherent and polychromatic incoherent imaging of a radial "star" target for equivalent t-number circular and Gaussian pupils. The FFT-based simulations are carried out using OpenMP threading on a multi-core desktop computer, with and without the aid of a many-core NVIDIA GPU accessing its cuFFT library. It is found that a custom FFT optimized for the 12-core host has similar performance to a simply implemented 256-core GPU FFT. A more sophisticated version of the latter but tuned to reduce overhead on a 448-core GPU is 20 to 28 times faster than a basic FFT implementation running on one CPU core.

A GPU-paralleled implementation of an enhanced face recognition algorithm

NASA Astrophysics Data System (ADS)

Chen, Hao; Liu, Xiyang; Shao, Shuai; Zan, Jiguo

2013-03-01

Face recognition algorithm based on compressed sensing and sparse representation is hotly argued in these years. The scheme of this algorithm increases recognition rate as well as anti-noise capability. However, the computational cost is expensive and has become a main restricting factor for real world applications. In this paper, we introduce a GPU-accelerated hybrid variant of face recognition algorithm named parallel face recognition algorithm (pFRA). We describe here how to carry out parallel optimization design to take full advantage of many-core structure of a GPU. The pFRA is tested and compared with several other implementations under different data sample size. Finally, Our pFRA, implemented with NVIDIA GPU and Computer Unified Device Architecture (CUDA) programming model, achieves a significant speedup over the traditional CPU implementations.

3D fully convolutional networks for subcortical segmentation in MRI: A large-scale study.

PubMed

Dolz, Jose; Desrosiers, Christian; Ben Ayed, Ismail

2018-04-15

This study investigates a 3D and fully convolutional neural network (CNN) for subcortical brain structure segmentation in MRI. 3D CNN architectures have been generally avoided due to their computational and memory requirements during inference. We address the problem via small kernels, allowing deeper architectures. We further model both local and global context by embedding intermediate-layer outputs in the final prediction, which encourages consistency between features extracted at different scales and embeds fine-grained information directly in the segmentation process. Our model is efficiently trained end-to-end on a graphics processing unit (GPU), in a single stage, exploiting the dense inference capabilities of fully CNNs. We performed comprehensive experiments over two publicly available datasets. First, we demonstrate a state-of-the-art performance on the ISBR dataset. Then, we report a large-scale multi-site evaluation over 1112 unregistered subject datasets acquired from 17 different sites (ABIDE dataset), with ages ranging from 7 to 64 years, showing that our method is robust to various acquisition protocols, demographics and clinical factors. Our method yielded segmentations that are highly consistent with a standard atlas-based approach, while running in a fraction of the time needed by atlas-based methods and avoiding registration/normalization steps. This makes it convenient for massive multi-site neuroanatomical imaging studies. To the best of our knowledge, our work is the first to study subcortical structure segmentation on such large-scale and heterogeneous data. Copyright © 2017 Elsevier Inc. All rights reserved.

Graphics processing unit (GPU) real-time infrared scene generation

NASA Astrophysics Data System (ADS)

Christie, Chad L.; Gouthas, Efthimios (Themie); Williams, Owen M.

2007-04-01

VIRSuite, the GPU-based suite of software tools developed at DSTO for real-time infrared scene generation, is described. The tools include the painting of scene objects with radiometrically-associated colours, translucent object generation, polar plot validation and versatile scene generation. Special features include radiometric scaling within the GPU and the presence of zoom anti-aliasing at the core of VIRSuite. Extension of the zoom anti-aliasing construct to cover target embedding and the treatment of translucent objects is described.

High performance MRI simulations of motion on multi-GPU systems

PubMed Central

2014-01-01

Background MRI physics simulators have been developed in the past for optimizing imaging protocols and for training purposes. However, these simulators have only addressed motion within a limited scope. The purpose of this study was the incorporation of realistic motion, such as cardiac motion, respiratory motion and flow, within MRI simulations in a high performance multi-GPU environment. Methods Three different motion models were introduced in the Magnetic Resonance Imaging SIMULator (MRISIMUL) of this study: cardiac motion, respiratory motion and flow. Simulation of a simple Gradient Echo pulse sequence and a CINE pulse sequence on the corresponding anatomical model was performed. Myocardial tagging was also investigated. In pulse sequence design, software crushers were introduced to accommodate the long execution times in order to avoid spurious echoes formation. The displacement of the anatomical model isochromats was calculated within the Graphics Processing Unit (GPU) kernel for every timestep of the pulse sequence. Experiments that would allow simulation of custom anatomical and motion models were also performed. Last, simulations of motion with MRISIMUL on single-node and multi-node multi-GPU systems were examined. Results Gradient Echo and CINE images of the three motion models were produced and motion-related artifacts were demonstrated. The temporal evolution of the contractility of the heart was presented through the application of myocardial tagging. Better simulation performance and image quality were presented through the introduction of software crushers without the need to further increase the computational load and GPU resources. Last, MRISIMUL demonstrated an almost linear scalable performance with the increasing number of available GPU cards, in both single-node and multi-node multi-GPU computer systems. Conclusions MRISIMUL is the first MR physics simulator to have implemented motion with a 3D large computational load on a single computer multi-GPU configuration. The incorporation of realistic motion models, such as cardiac motion, respiratory motion and flow may benefit the design and optimization of existing or new MR pulse sequences, protocols and algorithms, which examine motion related MR applications. PMID:24996972

Working memory, perceptual priming, and the perception of hierarchical forms: opposite effects of priming and working memory without memory refreshing.

PubMed

Kim, Jeong-Im; Humphreys, Glyn W

2010-08-01

Previous research has shown that stimuli held in working memory (WM) can influence spatial attention. Using Navon stimuli, we explored whether and how items in WM affect the perception of visual targets at local and global levels in compound letters. Participants looked for a target letter presented at a local or global level while holding a regular block letter as a memory item. An effect of holding the target's identity in WM was found. When memory items and targets were the same, performance was better than in a neutral condition when the memory item did not appear in the hierarchical letter (a benefit from valid cuing). When the memory item matched the distractor in the hierarchical stimulus, performance was worse than in the neutral baseline (a cost on invalid trials). These effects were greatest when the WM cue matched the global level of the hierarchical stimulus, suggesting that WM biases attention to the global level of form. Interestingly, in a no-memory priming condition, target perception was faster in the invalid condition than in the neutral baseline, reversing the effect in the WM condition. A further control experiment ruled out the effects of WM being due to participants' refreshing their memory from the hierarchical stimulus display. The data show that information in WM biases the selection of hierarchical forms, whereas priming does not. Priming alters the perceptual processing of repeated stimuli without biasing attention.

A Survey of Techniques for Approximate Computing

DOE PAGES

Mittal, Sparsh

2016-03-18

Approximate computing trades off computation quality with the effort expended and as rising performance demands confront with plateauing resource budgets, approximate computing has become, not merely attractive, but even imperative. Here, we present a survey of techniques for approximate computing (AC). We discuss strategies for finding approximable program portions and monitoring output quality, techniques for using AC in different processing units (e.g., CPU, GPU and FPGA), processor components, memory technologies etc., and programming frameworks for AC. Moreover, we classify these techniques based on several key characteristics to emphasize their similarities and differences. Finally, the aim of this paper is tomore » provide insights to researchers into working of AC techniques and inspire more efforts in this area to make AC the mainstream computing approach in future systems.« less

3D fast adaptive correlation imaging for large-scale gravity data based on GPU computation

NASA Astrophysics Data System (ADS)

Chen, Z.; Meng, X.; Guo, L.; Liu, G.

2011-12-01

In recent years, large scale gravity data sets have been collected and employed to enhance gravity problem-solving abilities of tectonics studies in China. Aiming at the large scale data and the requirement of rapid interpretation, previous authors have carried out a lot of work, including the fast gradient module inversion and Euler deconvolution depth inversion ,3-D physical property inversion using stochastic subspaces and equivalent storage, fast inversion using wavelet transforms and a logarithmic barrier method. So it can be say that 3-D gravity inversion has been greatly improved in the last decade. Many authors added many different kinds of priori information and constraints to deal with nonuniqueness using models composed of a large number of contiguous cells of unknown property and obtained good results. However, due to long computation time, instability and other shortcomings, 3-D physical property inversion has not been widely applied to large-scale data yet. In order to achieve 3-D interpretation with high efficiency and precision for geological and ore bodies and obtain their subsurface distribution, there is an urgent need to find a fast and efficient inversion method for large scale gravity data. As an entirely new geophysical inversion method, 3D correlation has a rapid development thanks to the advantage of requiring no a priori information and demanding small amount of computer memory. This method was proposed to image the distribution of equivalent excess masses of anomalous geological bodies with high resolution both longitudinally and transversely. In order to tranform the equivalence excess masses into real density contrasts, we adopt the adaptive correlation imaging for gravity data. After each 3D correlation imaging, we change the equivalence into density contrasts according to the linear relationship, and then carry out forward gravity calculation for each rectangle cells. Next, we compare the forward gravity data with real data, and comtinue to perform 3D correlation imaging for the redisual gravity data. After several iterations, we can obtain a satisfactoy results. Newly developed general purpose computing technology from Nvidia GPU (Graphics Processing Unit) has been put into practice and received widespread attention in many areas. Based on the GPU programming mode and two parallel levels, five CPU loops for the main computation of 3D correlation imaging are converted into three loops in GPU kernel functions, thus achieving GPU/CPU collaborative computing. The two inner loops are defined as the dimensions of blocks and the three outer loops are defined as the dimensions of threads, thus realizing the double loop block calculation. Theoretical and real gravity data tests show that results are reliable and the computing time is greatly reduced. Acknowledgments We acknowledge the financial support of Sinoprobe project (201011039 and 201011049-03), the Fundamental Research Funds for the Central Universities (2010ZY26 and 2011PY0183), the National Natural Science Foundation of China (41074095) and the Open Project of State Key Laboratory of Geological Processes and Mineral Resources (GPMR0945).

Efficient and portable acceleration of quantum chemical many-body methods in mixed floating point precision using OpenACC compiler directives

NASA Astrophysics Data System (ADS)

Eriksen, Janus J.

2017-09-01

It is demonstrated how the non-proprietary OpenACC standard of compiler directives may be used to compactly and efficiently accelerate the rate-determining steps of two of the most routinely applied many-body methods of electronic structure theory, namely the second-order Møller-Plesset (MP2) model in its resolution-of-the-identity approximated form and the (T) triples correction to the coupled cluster singles and doubles model (CCSD(T)). By means of compute directives as well as the use of optimised device math libraries, the operations involved in the energy kernels have been ported to graphics processing unit (GPU) accelerators, and the associated data transfers correspondingly optimised to such a degree that the final implementations (using either double and/or single precision arithmetics) are capable of scaling to as large systems as allowed for by the capacity of the host central processing unit (CPU) main memory. The performance of the hybrid CPU/GPU implementations is assessed through calculations on test systems of alanine amino acid chains using one-electron basis sets of increasing size (ranging from double- to pentuple-ζ quality). For all but the smallest problem sizes of the present study, the optimised accelerated codes (using a single multi-core CPU host node in conjunction with six GPUs) are found to be capable of reducing the total time-to-solution by at least an order of magnitude over optimised, OpenMP-threaded CPU-only reference implementations.

Analytical modeling and feasibility study of a multi-GPU cloud-based server (MGCS) framework for non-voxel-based dose calculations.

PubMed

Neylon, J; Min, Y; Kupelian, P; Low, D A; Santhanam, A

2017-04-01

In this paper, a multi-GPU cloud-based server (MGCS) framework is presented for dose calculations, exploring the feasibility of remote computing power for parallelization and acceleration of computationally and time intensive radiotherapy tasks in moving toward online adaptive therapies. An analytical model was developed to estimate theoretical MGCS performance acceleration and intelligently determine workload distribution. Numerical studies were performed with a computing setup of 14 GPUs distributed over 4 servers interconnected by a 1 Gigabits per second (Gbps) network. Inter-process communication methods were optimized to facilitate resource distribution and minimize data transfers over the server interconnect. The analytically predicted computation time predicted matched experimentally observations within 1-5 %. MGCS performance approached a theoretical limit of acceleration proportional to the number of GPUs utilized when computational tasks far outweighed memory operations. The MGCS implementation reproduced ground-truth dose computations with negligible differences, by distributing the work among several processes and implemented optimization strategies. The results showed that a cloud-based computation engine was a feasible solution for enabling clinics to make use of fast dose calculations for advanced treatment planning and adaptive radiotherapy. The cloud-based system was able to exceed the performance of a local machine even for optimized calculations, and provided significant acceleration for computationally intensive tasks. Such a framework can provide access to advanced technology and computational methods to many clinics, providing an avenue for standardization across institutions without the requirements of purchasing, maintaining, and continually updating hardware.

A survey of techniques for architecting and managing GPU register file

DOE PAGES

Mittal, Sparsh

2016-04-07

To support their massively-multithreaded architecture, GPUs use very large register file (RF) which has a capacity higher than even L1 and L2 caches. In total contrast, traditional CPUs use tiny RF and much larger caches to optimize latency. Due to these differences, along with the crucial impact of RF in determining GPU performance, novel and intelligent techniques are required for managing GPU RF. In this paper, we survey the techniques for designing and managing GPU RF. We discuss techniques related to performance, energy and reliability aspects of RF. To emphasize the similarities and differences between the techniques, we classify themmore » along several parameters. Lastly, the aim of this paper is to synthesize the state-of-art developments in RF management and also stimulate further research in this area.« less

A GPU-Based Implementation of the Firefly Algorithm for Variable Selection in Multivariate Calibration Problems

PubMed Central

de Paula, Lauro C. M.; Soares, Anderson S.; de Lima, Telma W.; Delbem, Alexandre C. B.; Coelho, Clarimar J.; Filho, Arlindo R. G.

2014-01-01

Several variable selection algorithms in multivariate calibration can be accelerated using Graphics Processing Units (GPU). Among these algorithms, the Firefly Algorithm (FA) is a recent proposed metaheuristic that may be used for variable selection. This paper presents a GPU-based FA (FA-MLR) with multiobjective formulation for variable selection in multivariate calibration problems and compares it with some traditional sequential algorithms in the literature. The advantage of the proposed implementation is demonstrated in an example involving a relatively large number of variables. The results showed that the FA-MLR, in comparison with the traditional algorithms is a more suitable choice and a relevant contribution for the variable selection problem. Additionally, the results also demonstrated that the FA-MLR performed in a GPU can be five times faster than its sequential implementation. PMID:25493625

A GPU-Based Implementation of the Firefly Algorithm for Variable Selection in Multivariate Calibration Problems.

PubMed

de Paula, Lauro C M; Soares, Anderson S; de Lima, Telma W; Delbem, Alexandre C B; Coelho, Clarimar J; Filho, Arlindo R G

2014-01-01

Several variable selection algorithms in multivariate calibration can be accelerated using Graphics Processing Units (GPU). Among these algorithms, the Firefly Algorithm (FA) is a recent proposed metaheuristic that may be used for variable selection. This paper presents a GPU-based FA (FA-MLR) with multiobjective formulation for variable selection in multivariate calibration problems and compares it with some traditional sequential algorithms in the literature. The advantage of the proposed implementation is demonstrated in an example involving a relatively large number of variables. The results showed that the FA-MLR, in comparison with the traditional algorithms is a more suitable choice and a relevant contribution for the variable selection problem. Additionally, the results also demonstrated that the FA-MLR performed in a GPU can be five times faster than its sequential implementation.

Accelerating electron tomography reconstruction algorithm ICON with GPU.

PubMed

Chen, Yu; Wang, Zihao; Zhang, Jingrong; Li, Lun; Wan, Xiaohua; Sun, Fei; Zhang, Fa

2017-01-01

Electron tomography (ET) plays an important role in studying in situ cell ultrastructure in three-dimensional space. Due to limited tilt angles, ET reconstruction always suffers from the "missing wedge" problem. With a validation procedure, iterative compressed-sensing optimized NUFFT reconstruction (ICON) demonstrates its power in the restoration of validated missing information for low SNR biological ET dataset. However, the huge computational demand has become a major problem for the application of ICON. In this work, we analyzed the framework of ICON and classified the operations of major steps of ICON reconstruction into three types. Accordingly, we designed parallel strategies and implemented them on graphics processing units (GPU) to generate a parallel program ICON-GPU. With high accuracy, ICON-GPU has a great acceleration compared to its CPU version, up to 83.7×, greatly relieving ICON's dependence on computing resource.

Rapid earthquake detection through GPU-Based template matching

NASA Astrophysics Data System (ADS)

Mu, Dawei; Lee, En-Jui; Chen, Po

2017-12-01

The template-matching algorithm (TMA) has been widely adopted for improving the reliability of earthquake detection. The TMA is based on calculating the normalized cross-correlation coefficient (NCC) between a collection of selected template waveforms and the continuous waveform recordings of seismic instruments. In realistic applications, the computational cost of the TMA is much higher than that of traditional techniques. In this study, we provide an analysis of the TMA and show how the GPU architecture provides an almost ideal environment for accelerating the TMA and NCC-based pattern recognition algorithms in general. So far, our best-performing GPU code has achieved a speedup factor of more than 800 with respect to a common sequential CPU code. We demonstrate the performance of our GPU code using seismic waveform recordings from the ML 6.6 Meinong earthquake sequence in Taiwan.

A survey of techniques for architecting and managing GPU register file

DOE Office of Scientific and Technical Information (OSTI.GOV)

Mittal, Sparsh

To support their massively-multithreaded architecture, GPUs use very large register file (RF) which has a capacity higher than even L1 and L2 caches. In total contrast, traditional CPUs use tiny RF and much larger caches to optimize latency. Due to these differences, along with the crucial impact of RF in determining GPU performance, novel and intelligent techniques are required for managing GPU RF. In this paper, we survey the techniques for designing and managing GPU RF. We discuss techniques related to performance, energy and reliability aspects of RF. To emphasize the similarities and differences between the techniques, we classify themmore » along several parameters. Lastly, the aim of this paper is to synthesize the state-of-art developments in RF management and also stimulate further research in this area.« less