A parallel implementation of the cyclic multigrid method for solving a boundary value problem for Poisson's equation in R³ is discussed for a graphics processor unit using NVIDIA CUDA technology. The results obtained by the graphics processing unit is compared with the analytical solution for the Dirichlet problem and with the numerical CPU-solution for the Dirichlet-Neumann problem. The acceleration of the parallel NVIDIA GeForce 8800 GTX code compared to the AMD Athlon 64X2 4800+ serial code is found to be about 200 times for 1 000 000 discrete elements. Moreover, a 8-core workstation based on two Intel(R) Xeon(R) 2.33HHz CPUs is found to be slower by 40 times than that of the GPU code.

Keywords: Poisson's equation, cyclic multigrid method, parallel computing, graphics processing unit, CUDA technology

• PostScript (in Russian) (7.68MB)
• PDF (in Russian) (529KB)



• PostScript. zip (in Russian) (825KB)
• PDF. zip (in Russian) (415KB)