Search NASA⌕ Search

SEARCH · Search NASA

Results for “Graphics Processing Unit (GPU) Parallel Computing”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 181 records · Page 10

WARP

WARP, which can stand for ``Weaving All the Random Particles,'' is a three-dimensional (3D) continuous energy Monte Carlo neutron transport code developed at UC Berkeley to efficiently execute on NVIDIA graphics processing unit (GPU) platforms. WARP accelerates Monte Carlo simulations while preserving the benefits of using the Monte Carlo method, namely, that very few physical and geometrical simplifications are applied. WARP is able to calculate multiplication factors, neutron flux distributions (in both space and energy), and fission source distributions for time-independent neutron transport problems. It can run in both criticality or fixed source modes, but fixed source mode is currently not robust, optimized, or maintained in the newest version. WARP can transport neutrons in unrestricted arrangements of parallelepipeds, hexagonal prisms, cylinders, and spheres. The goal of developing WARP is to investigate algorithms that can grow into a full-featured, continuous energy, Monte Carlo neutron transport code that is accelerated by running on GPUs. The crux of the effort is to make Monte Carlo calculations faster while producing accurate results. Modern supercomputers are commonly being built with GPU coprocessor cards in their nodes to increase their computational efficiency and performance. GPUs execute efficiently on data-parallel problems, but most CPU codes, including those for Monte Carlo neutral particle transport, are predominantly task-parallel. WARP uses a data-parallel neutron transport algorithm to take advantage of the computing power GPUs offer.

Bergmann, Ryan M.↗

Comparisons of dosimetric accuracy and calculation time of ARCHER and MCNP5 codes for the Ir-192 brachytherapy case - Paper 1

In the last decade, the graphics processing unit (GPU) technology has emerged as a viable player in high-performance computing. A GPU card is designed to carry out thousands of single instructions simultaneously, showing unprecedented computing power when compared with the traditional CPU technology. It has been proved that some computationally intensive programs can be accelerated drastically on the GPUs including Monte Carlo simulations. Unfortunately, existing production Monte Carlo codes cannot be easily ported to the GPU environment. To address the needs, a new Monte Carlo code ARCHER (Accelerated Radiation-transport Computations in Heterogeneous Environment) is being developed at Rensselaer Polytechnic Institute, employing the latest GPU and the Intel Xeon Phi coprocessor technologies. We can execute the code which was written using C and C++ on both CPU and GPU platforms. The ARCHER GPU code was written in CUDA C 5.0 and was tested on the Tesla M2090. There have been a few studies testing the ARCHER GPU code on several different platforms. In particular, Su et al has tested the ARCHER for external-beam radiation treatment 20 MeV monoenergetic monodirectional parallel electron beam. The comparison of the calculation results between EGSnrc and ARCHERGPU was impressive. A remarkable 224- fold speed-up factor was obtained using a NVIDIA Tesla M2090 GPU compared to a CPU. In this study, we applied the ARCHER code to dose calculation for a relatively new procedure, called accelerated partial breast irradiation (APBI) brachytherapy. The APBI brachytherapy dose calculation was based on RPI-AF phantom which is a 3D heterogeneous patient model. We then compared the dosimetric accuracy and calculation times of ARCHER and MCNP5 to gain insight into the accuracy and speedup of the GPU-based ARCHER code. (authors)

61 RADIATION PROTECTION AND DOSIMETRY↗

Parallel accelerated vector similarity calculations for genomics applications

The surge in availability of genomic data holds promise for enabling determination of genetic causes of observed individual traits, with applications to problems such as discovery of the genetic roots of phenotypes, be they molecular phenotypes such as gene expression or metabolite concentrations, or complex phenotypes such as diseases. However, the growing sizes of these datasets and the quadratic, cubic or higher scaling characteristics of the relevant algorithms pose a serious computational challenge necessitating use of leadership scale computing. In this study we describe a new approach to performing vector similarity metrics calculations, suitable for parallel systems equipped with graphics processing units (GPUs) or Intel Xeon Phi processors. Our primary focus is the Proportional Similarity metric applied to Genome Wide Association Studies (GWAS) and Phenome Wide Association Studies (PheWAS). We describe the implementation of the algorithms on accelerated processors, methods used for eliminating redundant calculations due to symmetries, and techniques for efficient mapping of the calculations to many-node parallel systems. Finally, results are presented demonstrating high per-node performance and parallel scalability with rates of more than five quadrillion (5 × 10 15 ) elementwise comparisons achieved per second on the ORNL Titan system. In a companion paper we describe corresponding techniques applied to calculations of the Custom Correlation Coefficient for comparative genomics applications.

59 BASIC BIOLOGICAL SCIENCES↗

A two-level GPU-accelerated incomplete LU preconditioner for general sparse linear systems

This paper presents a parallel preconditioning approach based on incomplete LU (ILU) factorizations in the framework of Domain Decomposition (DD) for general sparse linear systems. We focus on distributed memory parallel architectures, specifically, those that are equipped with graphic processing units (GPUs). In addition to block-Jacobi, we present general purpose two-level ILU Schur complement-based approaches, where different strategies are presented to solve the coarse-level reduced system. These strategies are combined with modified ILU methods in the construction of the coarse-level operator, in order to effectively remove smooth errors by targeting an algebraically smooth vector. We leverage available GPU-based sparse matrix kernels to accelerate the setup and the solve phases of the proposed ILU preconditioner. We evaluate the efficiency of the proposed methods as a smoother for algebraic multigrid (AMG) and as a preconditioner for Krylov subspace methods on challenging anisotropic diffusion problems and a collection of general sparse matrices.

97 MATHEMATICS AND COMPUTING↗

Investigation of Portable Event-Based Monte Carlo Transport Using the NVIDIA Thrust Library

Power consumption considerations are driving future high performance computing platforms toward many-core computing architectures. Los Alamos National Laboratory's Trinity machine, available in 2016, will use both Intel Xeon Haswell processors and Intel Xeon Phi Knights Landing many integrated core (MIC) architecture coprocessors. Lawrence Livermore National Laboratory's Sierra machine, available in 2018, will use an IBM PowerPC architecture along with Nvidia graphics processing unit (GPU) architecture accelerators. These different advanced architectures make the computing landscape in upcoming years complex. Traditional approaches to Monte Carlo transport do not work efficiently on these new computing platforms. MIC architectures require vectorization to operate efficiently, and vectorization is difficult to achieve in Monte Carlo transport. GPU architectures require additional code to explicitly use the hardware, requiring significant code changes or hardware specific branches in the source code. A significant challenge for Monte Carlo transport projects is to simultaneously support within a single source code base efficient simulations for both the current generation of architectures and the different advanced computing architectures. In order to address these challenges, two important changes are typically required: a new algorithmic approach for solving Monte Carlo transport, and explicit use of hardware specific software. In this paper, we describe initial research investigations of an event-based Monte Carlo transport algorithm implemented using the Nvidia Thrust library on a GPU for a Monte Carlo test code. The event-based algorithm targets many-core architectures by increasing SIMD (single instruction multiple data) parallelism, while Thrust potentially provides portable performance by allowing one source code base to compile code targeted for both CPUs and GPUs. We described preliminary investigations of portable event-based Monte Carlo algorithms implemented using the Nvidia Thrust library in a research Monte Carlo test code. We found that an explicit CUDA implementation of an event-based Monte Carlo algorithm performed significantly more efficiently than a Thrust implementation on GPU platforms, most likely as a result of additional flexibility in access to different memory spaces on the GPU. Additionally, we showed that on GPU platforms and at large enough problem sizes the event-based implementations perform more efficiently than the serial history-based implementation running on the host CPU. While investigating this problem, we also discovered that the performance of the event-based algorithm is affected by what tallies are being used. A zonal scalar flux tally requires atomic operations that significantly impacted the performance of the code, in some cases producing slowdowns instead of speedups. We decided to remove the tally in order to focus on the effectiveness of the event-based algorithm. Future work will be required to research more effective ways of handling such tallies. Additionally, we would like to consider new ways for optimizing both the Thrust and CUDA versions, in order to see how much performance we have yet to achieve. The potential trade-off between portability and performance was demonstrated in this investigation. Thrust provides both CPU and GPU versions of the code in one code base, but it does so at a cost. We discovered approximately a factor of 2-5 performance difference between Thrust and CUDA on each GPU platform we tested. Thrust provides a tool to access the GPUs with less effort and specialization, but it does so by giving up fine grained control where extra performance can be found for this application. Future work will be required to determine whether this performance differential between Thrust and CUDA can be reduced for event-based Monte Carlo transport. (authors)

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Parallelization of a Six Degree of Freedom Entry Vehicle Trajectory Simulation Using OpenMP and OpenACC

The art and science of writing parallelized software, using methods such as Open Multi-Processing (OpenMP) and Open Accelerators (OpenACC), is dominated by computer scientists. Engineers and non-computer scientists looking to apply these techniques to their project applications face a steep learning curve, especially when looking to adapt their original single threaded software to run multi-threaded on graphics processing units (GPUs). There are significant changes in mindset that must occur; such as how to manage memory, the organization of instructions, and the use of if statements (also known as branching). The purpose of this work is twofold: 1) to demonstrate the applicability of parallelized coding methodologies, OpenMP and OpenACC, to tasks outside of the typical large scale matrix mathematics; and 2) to discuss, from an engineer’s perspective, the lessons learned from parallelizing software using these computer science techniques. This work applies OpenMP, on both multi-core central processing units (CPUs) and Intel® Xeon Phi™ 7210, and OpenACC on GPUs. These parallelization techniques are used to tackle the simulation of thousands of entry vehicle trajectories through the integration of six degree of freedom (DoF) equations of motion (EoM). The forces and moments acting on the entry vehicle, and used by the EoM, are estimated using multiple models of varying levels of complexity. Several benchmark comparisons are made on the execution of six DoF trajectory simulation: single thread Intel® Xeon® E5-2670 CPU, multi-thread CPU using OpenMP, multi-thread Xeon Phi™ 7210 using OpenMP, and multi-thread NVIDIA® Tesla® K40 GPU using OpenACC. These benchmarks are run on the Pleiades Supercomputer Cluster at the National Aeronautics and Space Administration (NASA) Ames Research Center (ARC), and a Xeon Phi™ 7210 node at NASA Langley Research Center (LaRC).

Green, Justin S.↗

GIGA-Lens: Fast Bayesian Inference for Strong Gravitational Lens Modeling

We present GIGA-Lens: a gradient-informed, GPU-accelerated Bayesian framework for modeling strong gravitational lensing systems, implemented in TensorFlow and JAX. The three components, optimization using multistart gradient descent, posterior covariance estimation with variational inference, and sampling via Hamiltonian Monte Carlo, all take advantage of gradient information through automatic differentiation and massive parallelization on graphics processing units (GPUs). We test our pipeline on a large set of simulated systems and demonstrate in detail its high level of performance. The average time to model a single system on four Nvidia A100 GPUs is 105 s. The robustness, speed, and scalability offered by this framework make it possible to model the large number of strong lenses found in current surveys and present a very promising prospect for the modeling of ${ \mathcal O }({10}^{5})$ lensing systems expected to be discovered in the era of the Vera C. Rubin Observatory, Euclid, and the Nancy Grace Roman Space Telescope.

79 ASTRONOMY AND ASTROPHYSICS↗

Scaling soft matter physics to thousands of graphics processing units in parallel

We describe a multi-graphics processing unit (GPU) implementation of the Ludwig application, which specialises in simulating a variety of complex fluids via lattice Boltzmann fluid dynamics coupled to additional physics describing complex fluid constituents. We describe our methodology in augmenting the original central processing unit (CPU) version with GPU functionality in a maintainable fashion. We present several optimisations that maximise performance on the GPU architecture through tuning for the GPU memory hierarchy. We describe how we implement particles within the fluid in such a way to avoid a major diversion of the CPU and GPU codebases, whilst minimising data transfer at each time step. We detail our halo-exchange communication phase for the code, which exploits overlapping to allow efficient parallel scaling to many GPUs. We present results showing that the application demonstrates excellent scaling to at least 8192 GPUs in parallel, the largest system tested at the time of writing. The GPU version (on NVIDIA K20X GPUs) is around 3.5–5 times faster that the CPU version (on fully utilised AMD Opteron 6274 16-core CPUs), comparing equal numbers of CPUs and GPUs.

Computer Science↗

Distributed memory, GPU accelerated Fock construction for hybrid, Gaussian basis density functional theory

With the growing reliance of modern supercomputers on accelerator-based architecture such a graphics processing units (GPUs), the development and optimization of electronic structure methods to exploit these massively parallel resources has become a recent priority. While significant strides have been made in the development GPU accelerated, distributed memory algorithms for many modern electronic structure methods, the primary focus of GPU development for Gaussian basis atomic orbital methods has been for shared memory systems with only a handful of examples pursing massive parallelism. Here in this work, we present a set of distributed memory algorithms for the evaluation of the Coulomb and exact exchange matrices for hybrid Kohn–Sham DFT with Gaussian basis sets via direct density-fitted (DF-J-Engine) and seminumerical (sn-K) methods, respectively. The absolute performance and strong scalability of the developed methods are demonstrated on systems ranging from a few hundred to over one thousand atoms using up to 128 NVIDIA A100 GPUs on the Perlmutter supercomputer.

97 MATHEMATICS AND COMPUTING↗

Leaky Buddies: Cross-Component Covert Channels on Integrated CPU-GPU Systems

Graphics Processing Units (GPUs) are ubiquitous components used across the range of today’s computing plat- forms, from phones and tablets, through personal computers, to high-end server class platforms. With the increasing importance of graphics and video workloads, recent processors are shipped with GPU devices that are integrated on the same chip. Integrated GPUs share some resources with the CPU and as a result, there is a potential for microarchitectural attacks from the GPU to the CPU or vice versa. We consider the potential for covert channel attacks that arise either from shared microarchitectural components (such as caches) or through shared contention domains (e.g., shared buses). We illustrate these two types of channels by developing two reliable covert channel attacks. The first covert channel uses the shared LLC cache in Intel’s integrated GPU architectures. The second is a contention based channel targeting the ring bus connecting the CPU and GPU to the LLC. This is the first demonstrated microarchitectural attack crossing the component boundary (GPU to CPU or vice versa). Cross-component channels introduce a number of new challenges that we had to overcome since they occur across heterogeneous components that use different computation models and are interconnected using asymmetric memory hierarchies. We also exploit GPU parallelism to increase the bandwidth of the communication, even without relying on a common clock. The LLC based channel achieves a bandwidth of 120 kbps with a low error rate of 2%, while the contention based channel delivers up to 400 kbps with a 0.8% error rate. We also demonstrate a proof- of-concept prime-and-probe side channel attack that probes the full LLC from the GPU.

Cyber secruity, high performance comptuing, Hetero↗

Porting fragmentation methods to GPUs using an OpenMP API: Offloading the resolution-of-the-identity second-order Møller–Plesset perturbation method

Here, using an OpenMP Application Programming Interface, the resolution-of-the-identity second-order Møller–Plesset perturbation (RI-MP2) method has been off-loaded onto graphical processing units (GPUs), both as a standalone method in the GAMESS electronic structure program and as an electron correlation energy component in the effective fragment molecular orbital (EFMO) framework. First, a new scheme has been proposed to maximize data digestion on GPUs that subsequently linearizes data transfer from central processing units (CPUs) to GPUs. Second, the GAMESS Fortran code has been interfaced with GPU numerical libraries (e.g., NVIDIA cuBLAS and cuSOLVER) for efficient matrix operations (e.g., matrix multiplication, matrix decomposition, and matrix inversion). The standalone GPU RI-MP2 code shows an increasing speedup of up to 7.5× using one NVIDIA V100 GPU with one IBM 42-core P9 CPU for calculations on fullerenes of increasing size from 40 to 260 carbon atoms using the 6-31G(d)/cc-pVDZ-RI basis sets. A single Summit node with six V100s can compute the RI-MP2 correlation energy of a cluster of 175 water molecules using the correlation consistent basis sets cc-pVDZ/cc-pVDZ-RI containing 4375 atomic orbitals and 14 700 auxiliary basis functions in ~0.85 h. In the EFMO framework, the GPU RI-MP2 component shows near linear scaling for a large number of V100s when computing the energy of an 1800-atom mesoporous silica nanoparticle in a bath of 4000 water molecules. The parallel efficiencies of the GPU RI-MP2 component with 2304 and 4608 V100s are 98.0% and 96.1%, respectively.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Parallel accelerated Custom Correlation Coefficient calculations for genomics applications

The massive quantities of genomic data being made available through gene sequencing techniques are enabling breakthroughs in genomic science in many areas such as medical advances in the diagnosis and treatment of diseases. Analyzing this data, however, is a computational challenge insofar as the computational costs of the relevant algorithms can grow with quadratic, cubic or higher complexity—leading to the need for leadership scale computing. In this paper we describe a new approach to calculations of the Custom Correlation Coefficient (CCC) between Single Nucleotide Polymorphisms (SNPs) across a population, suitable for parallel systems equipped with graphics processing units (GPUs) or Intel Xeon Phi processors. We describe the mapping of the algorithms to accelerated processors, techniques used for eliminating redundant calculations due to symmetries, and strategies for efficient mapping of the calculations to many-node parallel systems. Results are presented demonstrating high per-node performance and near-ideal parallel scalability with rates of more than nine quadrillion (9 × 10 15 ) elementwise comparisons achieved per second with the latest optimized code on the ORNL Titan system, this being orders of magnitude faster than rates achieved using other codes and platforms as reported in the literature. Also it is estimated that as many as 90 quadrillion (90 × 10 15 ) comparisons per second may be achievable on the upcoming ORNL Summit system, an additional 10X performance increase. Finally, in a companion paper we describe corresponding techniques applied to calculations of the Proportional Similarity metric for comparative genomics applications.

97 MATHEMATICS AND COMPUTING↗

Gyrokinetic particle-in-cell optimization on emerging multi- and manycore platforms

The next decade of high-performance computing (HPC) systems will see a rapid evolution and divergence of multi- and manycore architectures as power and cooling constraints limit increases in microprocessor clock speeds. Understanding efficient optimization methodologies on diverse multicore designs in the context of demanding numerical methods is one of the greatest challenges faced today by the HPC community. In this paper, we examine the efficient multicore optimization of GTC, a petascale gyrokinetic toroidal fusion code for studying plasma microturbulence in tokamak devices. For GTC’s key computational components (charge deposition and particle push), we explore efficient parallelization strategies across a broad range of emerging multicore designs, including the recently-released Intel Nehalem-EX, the AMD Opteron Istanbul, and the highly multithreaded Sun UltraSparc T2+. We also present the first study on tuning gyrokinetic particle-in-cell (PIC) algorithms for graphics processors, using the NVIDIA C2050 (Fermi). Our work discusses several novel optimization approaches for gyrokinetic PIC, including mixed-precision computation, particle binning and decomposition strategies, grid replication, SIMDized atomic floating-point operations, and effective GPU texture memory utilization. Overall, we achieve significant performance improvements of 1.3–4.7× on these complex PIC kernels, despite the inherent challenges of data dependency and locality. Finally, our work also points to several architectural and programming features that could significantly enhance PIC performance and productivity on next-generation architectures.

97 MATHEMATICS AND COMPUTING↗

Accelerating gravitational microlensing simulations using the Xeon Phi coprocessor

Recently Graphics Processing Units (GPUs) have been used to speed up very CPU-intensive gravitational microlensing simulations. In this work, we use the Xeon Phi coprocessor to accelerate such simulations and compare its performance on a microlensing code with that of NVIDIA's GPUs. For the selected set of parameters evaluated in our experiment, we find that the speedup by Intel's Knights Corner coprocessor is comparable to that by NVIDIA's Fermi family of GPUs with compute capability 2.0, but less significant than GPUs with higher compute capabilities such as the Kepler. However, the very recently released second generation Xeon Phi, Knights Landing, is about 5.8 times faster than the Knights Corner, and about 2.9 times faster than the Kepler GPU used in our simulations. Here, we conclude that the Xeon Phi is a very promising alternative to GPUs for modern high performance microlensing simulations. LESS

79 ASTRONOMY AND ASTROPHYSICS↗

Algorithmic Patterns for $$\mathcal {H}$$-Matrices on Many-Core Processors

In this work, we consider the reformulation of hierarchical ( \(\mathcal {H}\) ) matrix algorithms for many-core processors with a model implementation on graphics processing units (GPUs). \(\mathcal {H}\) matrices approximate specific dense matrices, e.g., from discretized integral equations or kernel ridge regression, leading to log-linear time complexity in dense matrix–vector products. The parallelization of \(\mathcal {H}\) matrix operations on many-core processors is difficult due to the complex nature of the underlying algorithms. While previous algorithmic advances for many-core hardware focused on accelerating existing \(\mathcal {H}\) matrix CPU implementations by many-core processors, we here aim at totally relying on that processor type. As main contribution, we introduce the necessary parallel algorithmic patterns allowing to map the full \(\mathcal {H}\) matrix construction and the fast matrix–vector product to many-core hardware. In this work, crucial ingredients are space filling curves, parallel tree traversal and batching of linear algebra operations. The resulting model GPU implementation hmglib is the, to the best of the authors knowledge, first entirely GPU-based Open Source \(\mathcal {H}\) matrix library of this kind. We investigate application examples as present in kernel ridge regression, Gaussian Process Regression and kernel-based interpolation. In this context, an in-depth performance analysis and a comparative performance study against a standard multi-core CPU \(\mathcal {H}\) matrix library highlights profound speedups of our many-core parallel approach.

97 MATHEMATICS AND COMPUTING↗

GPU/MIC Acceleration of the LHC High Level Trigger to Extend the Physics Reach at the LHC

The quest for rare new physics phenomena leads the PI to propose evaluation of coprocessors based on Graphics Processing Units (GPUs) and the Intel Many Integrated Core (MIC) architecture for integration into the trigger system at LHC. This will require development of a new massively parallel implementation of the well known Combinatorial Track Finder which uses the Kalman Filter to accelerate processing of data from the silicon pixel and microstrip detectors and reconstruct the trajectory of all charged particles down to momentums of 100 MeV. It is expected to run at least one order of magnitude faster than an equivalent algorithm on a quad core CPU for extreme pileup scenarios of 100 interactions per bunch crossing. The new tracking algorithms will be developed and optimized separately on the GPU and Intel MIC and then evaluated against each other for performance and power efficiency. The results will be used to project the cost of the proposed hardware architectures for the HLT server farm, taking into account the long term projections of the main vendors in the market (AMD, Intel, and NVIDIA) over the next 10 years. Extensive experience and familiarity of the PI with the LHC tracker and trigger requirements led to the development of a complementary tracking algorithm that is described in [arxiv: 1305.4855], [arxiv: 1309.6275] and preliminary results accepted to JINST.

97 MATHEMATICS AND COMPUTING↗

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

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 partitions 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).

97 MATHEMATICS AND COMPUTING↗

Compressed basis GMRES on high-performance graphics processing units

Krylov methods provide a fast and highly parallel numerical tool for the iterative solution of many large-scale sparse linear systems. To a large extent, the performance of practical realizations of these methods is constrained by the communication bandwidth in current computer architectures, motivating the investigation of sophisticated techniques to avoid, reduce, and/or hide the message-passing costs (in distributed platforms) and the memory accesses (in all architectures). This article leverages Ginkgo’s memory accessor in order to integrate a communication-reduction strategy into the (Krylov) GMRES solver that decouples the storage format (i.e., the data representation in memory) of the orthogonal basis from the arithmetic precision that is employed during the operations with that basis. Given that the execution time of the GMRES solver is largely determined by the memory accesses, the cost of the datatype transforms can be mostly hidden, resulting in the acceleration of the iterative step via a decrease in the volume of bits being retrieved from memory. Together with the special properties of the orthonormal basis (whose elements are all bounded by 1), this paves the road toward the aggressive customization of the storage format, which includes some floating-point as well as fixed-point formats with mild impact on the convergence of the iterative process. We develop a high-performance implementation of the “compressed basis GMRES” solver in the Ginkgo sparse linear algebra library using a large set of test problems from the SuiteSparse Matrix Collection. We demonstrate robustness and performance advantages on a modern NVIDIA V100 graphics processing unit (GPU) of up to 50% over the standard GMRES solver that stores all data in IEEE double-precision.

97 MATHEMATICS AND COMPUTING↗