Parallelism of MRT Lattice Boltzmann Method based on Multi-GPUs
L. Yao, A. Wang, X. Ji
In order to accelerate the Multi-Relaxation-Time Lattice Boltzmann Method (MRT-LBM), it was parallelized on multi-GPUs located in one computer. Lattices of the MRT-LBM were distributed to the different GPUs and they were operated on the GPU threads concurrently. In
the step streaming, in order to get the neighbour lattice nformation, all lattices were combined and operated on the master GPU. Three-dimensional (3D) groundwater flow was simulated by
using the parallelized MRT-LBM and the experimental results showed that almost 95 times acceleration was attained on a six GPUs computer. The parallelized MRT-LBM based on the multi-GPUs located in one computer is efficient in terms of both time and energy.