GPU-Based Implementation of Finite Element Method for Elasticity Using CUDA

Graphics Processing Unit (GPU) has obtained great success in scientific computations for its tremendous computational horsepower and very high memory bandwidth. This paper discusses the way to accelerate the finite element method (FEM) for elasticity problem on NVIDIA GPUs using Compute Unified Device Architecture (CUDA), mainly including formation and solution of finite element equations. Multiple strategies for efficiently accessing global memory are introduced to achieve memory coalescing. Shared memory is utilized to reuse data for enhancing memory bandwidth efficiency. The numerical results show that GPU-accelerated finite element computation with using these optimizing strategies can achieve a speedup of around 10-20x to serial version for different element types.