high performance computing on graphics processing units: hgpu.org

This paper deals with a research concept of parallel finite element (FE) simulation for moving boundary and adaptive refinement problems using graphics processing unit (GPU). The main concern in this study is to improve the numerical performance of continuous FE simulation using recent data-parallel computing technology (GPU-CUDA). The computational time for our existing simulations is very long using conventional parallel computing technique (MPI). This short-come can be overcome using data parallel computing power of CPU and GPU by increasing the overall performance of FE simulation. By adapting the computing power of graphic processors for multi-threaded fine-grain parallelization for FE assembly and solving, overall performance can be significantly improved. This study will particularly concentrated on dynamic FE simulation of moving boundary and adaptively refining mesh domain.