Read the abstract, see who wrote it and who cites it, and get the citation in APA, MLA, Chicago or Harvard.
About 2,482 papers for “multi-GPU computing”
Chathika Gunaratne, Zhang, Xi, Prasanna Date et al. · OSTI OAI (U.S. Department of Energy Office of Scientific and Technical Information) · 2023
A GPU-based multi-agent simulation framework for neuromorphic computing
Cited by 0Open access
H I Alzeini, Sh A Hameed, Mohamed Hadi Habaebi · IOP Conference Series Materials Science and Engineering · 2013
The overwhelmingly increasing amount of stored data has spurred researchers seeking different methods in order to optimally take advantage of it which mostly have faced a response time problem as a result of this enormous size of data. Most of solutions have suggested materialization as a favourite solution. However, such a solution cannot attain Real-Time answers anyhow. In this paper we propose a framework illustrating the barriers and suggested solutions in the way of achieving Real-Time OLAP answers that are significantly used in decision support systems and data warehouses.
Cited by 1Open access
Song Jun Park · 2009
Parallel systems are becoming ubiquitous in the world of computing as evidenced by multi-core processors, heterogeneous Cell broadband engine, and highly parallel graphics processing units (GPUs). All parallel systems share a requirement that parallel programming is necessary to leverage multiple cores. As a result of this trend, multi-core CPUs are no longer a clear winner due to its peaked clock frequency and programming effort involved in parallelizing code for multi-core architecture. Given such drawbacks, dataparallel applications might benefit from GPU assisted computing. GPUs are the most popular and inexpensive accelerators. To evaluate GPU-based computing, a floating-point intensive algorithm for a radar imaging application is chosen for analysis. The paper attempts to present a fair performance comparison of CPU and GPU implementations.
Cited by 9
Jayanth Channagiri, Alexandre Boulle, A. Debelle · Acta Crystallographica Section A Foundations and Advances · 2015
Cited by 0Open access
Haoyu RU, Peixuan ZOU, Yufei LIU et al. · Chinese Journal of Aeronautics · 2026
The Discontinuous Galerkin (DG) method combines the strengths of the finite volume and finite element methods, making it ideal for Graphics Processing Unit (GPU) parallel computing. This paper presents a multi-GPU parallel algorithm for the DG method in the simulation of acoustic propagation governed by linearized Euler equations on a three-dimensional unstructured grid, using the Message Passing Interface (MPI) and the Compute Unified Device Architecture (CUDA). Several key memory access optimization strategies are adopted to leverage GPU hardware characteristics, including restructuring data to improve kernel function access efficiency to global memory and utilizing GPU high-speed memory to reduce global memory accesses. To ensure the accuracy and efficiency of parallel computing, a store-then-add strategy is adopted to effectively avoid memory access conflicts, and kernel functions are split to prevent warp divergence. Roofline model tests demonstrate that the performance of the kernel functions is close to the theoretical upper limit. For single-GPU parallel computing, the solver’s accuracy and high performance are validated using a Gaussian pulse example. This achieves approximately 400 times acceleration in the third- and fourth-order accuracy for three-dimensional unstructured grid computations. In multi-GPU parallel computing, MPI and CUDA streams overlap computation and communication. Experimental results show that the strong scalability parallel efficiency of 8 A800 GPUs reaches 81.32%, and the weak scalability parallel efficiency reaches 98.20%. These results demonstrate the excellent performance and scalability of the multi-GPU parallel algorithm implemented for large-scale parallel computing.
Cited by 0Open access
Manuel Costanzo, Enzo Rucci, Carlos García et al. · Future Generation Computer Systems · 2025
The high-performance computing (HPC) landscape is undergoing rapid transformation, with an increasing emphasis on energy-efficient and heterogeneous computing environments. This comprehensive study extends our previous research on SYCL’s performance portability by evaluating its effectiveness across a broader spectrum of computing architectures , including CPUs, GPUs , and hybrid CPU–GPU configurations from NVIDIA, Intel, and AMD. Our analysis covers single-GPU, multi-GPU, single-CPU, and CPU–GPU hybrid setups, using two common, bioinformatic applications as a case study . The results demonstrate SYCL’s versatility across different architectures, maintaining comparable performance to CUDA on NVIDIA GPUs while achieving similar architectural efficiency rates on AMD and Intel GPUs in the majority of cases tested. SYCL also demonstrated remarkable versatility and effectiveness across CPUs from various manufacturers, including the latest hybrid architectures from Intel. Although SYCL showed excellent functional portability in hybrid CPU–GPU configurations, performance varied significantly based on specific hardware combinations. Some performance limitations were identified in multi-GPU and CPU–GPU configurations, primarily attributed to workload distribution strategies rather than SYCL-specific constraints. These findings position SYCL as a promising unified programming model for heterogeneous computing environments, particularly for bioinformatic applications. • SYCL’s performance portability was analyzed on diverse, multiple-vendor architectures. • SYCL reached comparable performance to CUDA on NVIDIA GPUs. • SYCL achieved higher efficiency rates in most test cases from the other GPU vendors. • SYCL obtained similar architectural efficiency rates on most CPU configurations. • SYCL showed excellent functional portability in multi-GPU and CPU–GPU configurations. • SYCL stands as a promising unified programming model for heterogeneous systems.
Cited by 4Open access
Manuel Costanzo, Enzo Rucci, Carlos García et al. · arXiv (Cornell University) · 2024
The high-performance computing (HPC) landscape is undergoing rapid transformation, with an increasing emphasis on energy-efficient and heterogeneous computing environments. This comprehensive study extends our previous research on SYCL's performance portability by evaluating its effectiveness across a broader spectrum of computing architectures, including CPUs, GPUs, and hybrid CPU-GPU configurations from NVIDIA, Intel, and AMD. Our analysis covers single-GPU, multi-GPU, single-CPU, and CPU-GPU hybrid setups, using two common, bioinformatic applications as a case study. The results demonstrate SYCL's versatility across different architectures, maintaining comparable performance to CUDA on NVIDIA GPUs while achieving similar architectural efficiency rates on AMD and Intel GPUs in the majority of cases tested. SYCL also demonstrated remarkable versatility and effectiveness across CPUs from various manufacturers, including the latest hybrid architectures from Intel. Although SYCL showed excellent functional portability in hybrid CPU-GPU configurations, performance varied significantly based on specific hardware combinations. Some performance limitations were identified in multi-GPU and CPU-GPU configurations, primarily attributed to workload distribution strategies rather than SYCL-specific constraints. These findings position SYCL as a promising unified programming model for heterogeneous computing environments, particularly for bioinformatic applications.
Cited by 0Open access
Wangda Zuo, Qingyan Chen · 2009
Computational fluid dynamics (CFD) can provide detailed information of flow motion, temperature distributions and species dispersion in buildings. However, it may take hours or days, even weeks to simulate airflow in a building by using CFD on a single central processing unit (CPU) computer. Parallel computing on a multi-CPU supercomputer or computer cluster can reduce the computing time, but the cost for such high performance computing is prohibitive for many designers. Our paper introduces high performance parallel computing of the airflow simulations on a graphics processing unit (GPU). The computing time can be reduced by 10- 30 times using the GPU. Furthermore, the cost of purchasing such a GPU is only $500, which is less than 2 % of a multi-CPU supercomputer or a computer cluster for the same performance.
Cited by 2
2015
......As we enter the era of GPU computing, demanding applications with substantial parallelism increasingly use the massively parallel computing capabilities of GPUs to achieve superior performance and efficiency. Today GPU computing enables applications that we previously thought infeasible because of long execu-tion times. With the GPU’s rapid evolution from a configurable graphics processor to a pro-grammable parallel processor, the ubiqui-tous GPU in every PC, laptop, desktop, and workstation is a many-core multi-threaded multiprocessor that excels at
Cited by 0
Jianqi Lai, Fengshun Lu, Xingzhi Hu et al. · Journal of Physics Conference Series · 2024
Abstract To achieve a large-scale and efficient numerical solution to supersonic combustion chemical reaction problems, a numerical simulation algorithm for combustion chemical reactions was established on a graphics processing unit (GPU). Numerical simulation of the combustion chemical reaction of a hydrocarbon fuel/air mixture in a supersonic combustor was conducted to verify the accuracy of GPU parallel calculation results and analyze its parallel performance. Numerical results show that GPU parallel computing can accurately simulate the complex flowfield in the combustor. The distribution law of wall pressure is basically consistent with the experimental values, and the numerical agreement is good, indicating that the numerical simulation algorithm for the multi-GPU supersonic combustion chemical reaction established in this paper is correct and reliable. On the Tesla V100 GPU parallel computing platform, the speedup of GPU parallel computing can reach over 100 times, greatly improving computational efficiency. When the grid size is 18.64 million, the parallel efficiency of single and double precision floating-point operations on four GPUs is 66.2% and 69.1%, respectively, which demonstrates that the multi-GPU parallel algorithm has good scalability.
Cited by 0Open access
Meinke, Jan, Herten, Andreas, Hrywniak, Markus et al. · JuSER Publikationsportal · 2024
GPU-accelerated computing drives current scientific research. Writing fast numeric algorithms for GPUs offers high application performance by offloading compute-intensive portions of the code to a GPU.This advanced course consists of modules providing more in-depth coverage of multi-GPU programming, modern CUDA concepts, CUDA Fortran, and portable programming models such as OpenACC and C++ parallel STL algorithms.Topics covered will includeA) Advanced Multi-GPU Programming with MPIB) Advanced Multi-GPU Programming with NCCL and NVSHMEMC) Advanced and Modern CUDA Concepts (Cooperative Groups, CUDA Graphs, CUB Primitives, Modern C++ Programming)D) CUDA FortranE) GPU Programming with Abstractions (OpenACC, Standard Language Programming (pSTL))
Cited by 0Open access
Xvlong Wei, Shuqiang Chen · 2012
The rigorous coupled-wave analysis of two-dimensional gratings is revived. Parallel computing method is introduced in the codes based on RCWA. Hybrid computation using GPU with CPU is used to improve the efficiency of the RCWA simulation program, There are combination of highly parallel computing ability of GPU, and the fast serial computing ability of CPU. a typical calculation for 2-D grating shows that the computing speed can be accelerated by using the parallel computing method. the test used two CPU based math libraries-LAPACK, MKL, and the GPU code we have built. in this test, both single process and multi process are used in the same platform.
Cited by 1
Weicheng Xue, Christopher J. Roy · AIAA Scitech 2020 Forum · 2020
This paper deals with the CPU-GPU heterogeneous computing of a multi-block structured grid CFD code using multiple CPUs and GPUs. First, a brief description on how the heterogeneous computing framework is implemented using MPI and OpenACC is given. OpenACC directives are added for GPUs so most of the CFD code base is the same for CPUs and GPUs except for some specific performance optimizations for the GPU which have no significant effect on the CPU performance. Then, different domain decomposition strategies and a domain aggregation method along with their advantages tailored for our research CFD code, our applications, and the accelerator used are presented. The speedup and scaling performance of the CPU-GPU heterogeneous computing and the comparison with multi-CPU/GPU performance for several test cases including a supersonic inlet, a NACA 0012 airfoil and an ONERA M6 wing are given. Using a GPU or two GPUs and a certain number of CPUs together, the program solver time can be reduced compared with using pure CPUs or GPUs. Finally, conclusions are drawn to provide 1) suggestions for programmers who have an interest in developing their own CPU-GPU heterogeneous computing framework and 2) feedback for hardware researchers who are working on the design of the future CPU-GPU heterogeneous systems.
Cited by 4
Xinyu Yang, Xin Wang, Lei Yan et al. · 2022
With the rapid development of Internet of Things technology, the requirements for edge node data processing capability are increasing, and GPU processors are becoming more widely applied in edge nodes. Current research on GPU virtualization technology mainly focuses on cloud data centers, with little research on embedded GPU virtualization in scenarios with limited edge node resources. In contrast to cloud data centers, embedded GPUs in edge nodes typically do not have access to GPU utilization and video memory usage within each container. As a result, traditional GPU virtualization technologies are ineffective for resource virtualization on embedded devices. This paper presents a method to virtualize embedded GPU resources in an edge computing environment, called sGPU. We integrated edge nodes with embedded GPUs into Kubernetes via the device-plugin mechanism. Users can package GPU applications in containers and deploy them using Kubernetes on edge nodes with embedded GPUs. sGPU allows containers to share embedded GPU computing resources and divides a physical GPU into multiple virtual GPUs that can be allocated to containers on demand. To achieve GPU computing power division, we proposed a multi-container sharing GPU algorithm and implemented it in sGPU, which ensures the most accurate computing power segmentation under the competition of computing resources of a GPU used by multiple containers at the same time. According to the experimental results, the average overhead of sGPU is 3.28%. The accuracy of computing power segmentation is 92.7% when a single container uses GPU.
Cited by 2
2014
The following topics are dealt with: computer architecture and high performance computing; multicore architecture; GPU computing; multi-core computing; data-flow programming models and machines; parallel and distributed computing for big data applications; and distributed processing and data management.
Cited by 0
Shirley Holt, Philip B. Deverall, Jean E. Boddy · The Journal of Pathology · 1978
We describe a parallel implementation of a compressible Lattice Boltzmann code on a multi-GPU cluster based on Nvidia Fermi processors. We analyze how to optimize the algorithm for GP-GPU architectures, describe the implementation choices that we have adopted and compare our performance results with an implementation optimized for latest generation multi-core CPUs. Our program runs at ˜¿30% of the double-precision peak performance of one GPU and shows almost linear scaling when run on the multi-GPU cluster.\nKeywords: Computational fluid-dynamics – Lattice Boltzmann methods – GP-GPUs computing
Cited by 34
Pablo Javier Vidal, Enrique Alba · 2010
In this paper, we present a novel implementation of a Cellular Genetic Algorithm (cGA) model for a multi-GPU platform using NVIDIA's CUDA technology. This multi-GPU cGA model is compared first against a serial version in CPU and then versus an implementation on a single GPU. We divide the different operations of the cGA into distinct sets of instructions called kernels. Using the multi-GPU platform we observe that the speedup with respect to the CPU version ranges from 8 to 771, while it is similar to that of the GPU, with a little overhead in the multi-GPU case. Our results demonstrate that multi-GPU desktops can serve as cost-effective parallel computing platforms to obtain accurate results in very short time, although they need special considerations in order to improve on regular single GPUs.
Cited by 36
Tomoaki Nagaoka, Soichi Watanabe · 2012
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.
Cited by 15
曹安宏; Tsao, An-Hung · 2014
碩士
Cited by 0
Carlos Agulló-Domingo, Óscar Vera-López, Seyda Guzelhan et al. · ArXiv.org · 2025
Word-wise Fully Homomorphic Encryption (FHE) schemes, such as CKKS, are gaining significant traction due to their ability to provide post-quantum-resistant, privacy-preserving approximate computing; an especially desirable feature in Machine-Learning-as-a-Service (MLaaS) cloud-computing paradigms. OpenFHE is a leading CPU-based FHE library with robust CKKS operations, but its server-side performance is not yet sufficient for practical cloud deployment. As GPU computing becomes more common in data centers, many FHE libraries are adding GPU support. However, integrating an efficient GPU backend into OpenFHE is challenging. While OpenFHE uses a Hardware Abstraction Layer (HAL), its flexible architecture sacrifices performance due to the abstraction layers required for multi-scheme and multi-backend compatibility. In this work, we introduce FIDESlib, the first open-source server-side CKKS GPU library that is fully interoperable with well-established client-side OpenFHE operations. Unlike other existing open-source GPU libraries, FIDESlib provides the first implementation featuring heavily optimized GPU kernels for all CKKS primitives, including bootstrapping. Our library also integrates robust benchmarking and testing, ensuring it remains adaptable to further optimization. Furthermore, its software architecture is designed to support extensions to a multi-GPU backend for enhanced acceleration. Our experiments across various GPU systems and the leading open-source CKKS library to date, Phantom, show that FIDESlib offers superior performance and scalability. For bootstrapping, FIDESlib achieves no less than 70x speedup over the AVX-optimized OpenFHE implementation.
Cited by 0Open access