Plasma and Fusion Research
Volume 6, 2401075 (2011)
Regular Articles
- Department of Information System Fundamentals, The University of Electro-Communications, Chofu 182-8585, Japan
- 1)
- Keihanna Research Laboratories, National Institute of Information and Communications Technology, Soraku-gun 619-0289, Japan
- 2)
- Department of Earth and Planetary Sciences, Kyushu University, Fukuoka 812-8581, Japan
- 3)
- National Institute of Information and Communications Technology, Koganei 184-8795, Japan
Abstract
Particle-in-cell (PIC) is a simulation technique for plasma physics. The large number of particles in high-resolution plasma simulation increases the volume computation required, making it vital to increase computation speed. In this study, we attempt to accelerate computation speed on graphics processing units (GPUs) using KEMPO, a PIC simulation code package [H. Matsumoto and Y. Omura, Computer Space Plasma Physics, pp.21-65 (1985)]. We perform two tests for benchmarking, with small and large grid sizes. In these tests, we run KEMPO1 code using a CPU only, both a CPU and a GPU, and a GPU only. The results showed that performance using only a GPU was twice that of using a CPU alone. While, execution time for using both a CPU and GPU is comparable to the tests with a CPU alone, because of the significant bottleneck in communication between the CPU and GPU.
Keywords
numerical simulation, particle-in-cell method, graphics processing units, high-performance computing, OpenMP
Full Text
References
- [1] H. Matsumoto and Y. Omura, Computer Space Plasma Physics, pp.21-65 (1985).
- [2] NDVIA CUDA programming guide 1.1, http://developer.download.nvidia.com/compute/cuda/1_1/NVIDIA_CUDA_Programming_Guide_1.1.pdf.
- [3] Y. Akiyama et al., Information Processing Society of Japan 66, 1 (1997).
- [4] G. Stantchev et al., Journal of Parallel and Distributed Computing 68, 1339 (2008).
This paper may be cited as follows:
Junya SUZUKI, Hironori SHIMAZU, Keiichiro FUKAZAWA and Mitsue DEN, Plasma Fusion Res. 6, 2401075 (2011).