With the continuous improvement and expansion of GPU in programmable ability,the ability of parallel processing and the scope of application,GPU has developed into a processor which has highly parallel,multi thread and multi core[1].Using the GPU parallel processing ability,the heterogeneous parallel computing system which is characterized by the mixed acceleration of CPU+GPU will become the mainstream of future high performance computing.For the accurate simulation of some complex models,the researchers must solve two questions that are long computation time and big occupied computer storage.According to the programmable ability of GPU and the ability of high speed parallel processing[2],this article makes CUDA[3] in both macro scale and micro scale' s numerical simulation of particle reinforced composite materials and this method can speed the parallel computing better.This article use CUDA code rewriting the program of the most time-consuming calculation,and the conclusion shows that the cuda based parallel computing is much more efficiency.And we can get a more than 22 times the acceleration.