Parallel Computing of Discrete Element Method on GPU

Teruyoshi Washizawa, Yasuhiro Nakahara

Simulation & Analysis R&D Center, Canon Inc., Tokyo, Japan.

DOI: 10.4236/am.2013.41A037                

Abstract

General purpose computing on GPU for scientific computing has been rapidly growing in recent years. We investigate the applicability of GPU to discrete element method (DEM) often used in particle motion simulation. NVIDIA provides a sample code for this type of simulation, which obtained superior performance than CPU in computational time. A computational model of the contact force in NVIDIA’s sample code is, however, too simple to use in practice. This paper modifies the NVIDIA’s simple model by replacing it with the practical model. The computing speed of the practical model on GPU is compared with the simple one on GPU and with the practical one on CPU in numerical experiments. The result shows that the practical model on GPU obtains the computing speed 6 times faster than the practical one on CPU while 7 times slower than that of the simple one on GPU. The effects of the GPU architectures on the computing speed are analyzed.

Keywords

GPU; Particle Motion Simulation; Discrete Element Method; Warp Divergence

Conflicts of Interest

The authors declare no conflicts of interest.

References

[1]	http://www.nvidia.com
[2]	http://www.nvidia.com/object/gpu-applications.html
[3]	Y. Shigeto and M. Sakai, “Parallel Computing of Discrete Element Method on Multi-Core Processors,” Particuology, Vol. 9, No. 4, 2011, pp. 398-405. doi:10.1016/j.partic.2011.04.002
[4]	D. Nishiura and H. Sakaguchi, “Parallel-Vector Algorithms for Particle Simulations on Shared-Memory Multiprocessors,” Journal of Computational Physics, Vol. 230, No. 5, 2011, pp. 1923-1938. doi:10.1016/j.jcp.2010.11.040
[5]	NVIDIA, “NVIDIA CUDATM Programming Guide Version 3.0,” NVIDIA Corporation, Santa Clara, 2010.
[6]	M. Sakai, “Numerical Simulation of Granular Flows,” Maruzen Co. Ltd., Tokyo, 2012, pp. 17-40.
[7]	P. Du, R. Weber, P. Luszczek, S. Tomov, G. Peterson and J. Dongarra, “From CUDA to OpenCL: Towards a Performance-Portable Solution for Multi-Platform GPU Programming,” Parallel Computing, Vol. 38, No. 8, 2012, pp. 391-407. doi:10.1016/j.parco.2011.10.002
[8]	Y. Nakahara and T. Washizawa, “Obstacles for Applying Discrete Element Method to GPU,” Proceedings of the JSIAM, Wakkanai, 29 August 2012, pp. 11-12.