Multi-Hybrid Accelerated Simulation by GPU and FPGA on Radiative Transfer Simulation in Astrophysics

Access this Article

Author(s)

    • Kobayashi Ryohei
    • Center for Computational Sciences, University of Tsukuba|Degree Programs in Systems and Information Engineering, University of Tsukuba
    • Fujita Norihisa
    • Center for Computational Sciences, University of Tsukuba|Degree Programs in Systems and Information Engineering, University of Tsukuba
    • Yamaguchi Yoshiki
    • Center for Computational Sciences, University of Tsukuba|Degree Programs in Systems and Information Engineering, University of Tsukuba
    • Boku Taisuke
    • Center for Computational Sciences, University of Tsukuba|Degree Programs in Systems and Information Engineering, University of Tsukuba
    • Yoshikawa Kohji
    • Center for Computational Sciences, University of Tsukuba|Degree Programs in Pure and Applied Sciences, University of Tsukuba
    • Abe Makito
    • Center for Computational Sciences, University of Tsukuba
    • Umemura Masayuki
    • Center for Computational Sciences, University of Tsukuba|Degree Programs in Pure and Applied Sciences, University of Tsukuba

Abstract

<p>Field-programmable gate arrays (FPGAs) have garnered significant interest in research on high-performance computing because their computation and communication capabilities have drastically improved in recent years due to advances in semiconductor integration technologies that rely on Moore's Law. In addition to improving FPGA performance, toolchains for the development of FPGAs in OpenCL have been developed and offered by FPGA vendors that reduce the programming effort required. These improvements reveal the possibility of implementing a concept to enable on-the-fly offloading computation at which CPUs/GPUs perform poorly to FPGAs while performing low-latency data movement. We think that this concept is key to improving the performance of heterogeneous supercomputers using accelerators such as the GPU. In this paper, we propose a GPU-FPGA-accelerated simulation based on the concept and show our implementation with CUDA and OpenCL mixed programming for the proposed method. The results of experiments show that our proposed method can always achieve a better performance than GPU-based implementation and we believe that realizing GPU-FPGA-accelerated simulation is the most significant difference between our work and previous studies.</p>

Journal

  • Journal of Information Processing

    Journal of Information Processing 28(0), 1073-1089, 2020

    Information Processing Society of Japan

Codes

Page Top