GPU accelerated program for grid generation around two dimensional airfoils

Abstract

The most time consuming part of CFD simulation is geometry modeling and grid generation especially when grid adaptation is needed. One can utilize GPUs to accelerate such a process. In the present study, this acceleration is performed for generating structure grid around arbitrary airfoils using Poisson equations. An iterative Compute Unified Device Architecture (CUDA) code is developed to solve the system of elliptic equations. The study shows that based on the comparison between the Intel Core 2 Duo (CPU) and NVIDIA GeForce 480 GTX (GPU), speedups up to 151x with double precision arithmetic can be obtained. In addition to speedup comparison, the GPU performance analysis is also carried and the study shows that as much as 55.6 percent of theoretical memory throughput usage can be achieved. Such an achievement shows that using GPU programming in grid generation process will accelerate grid adaptation and refinement considerably.