Formal Description and Optimization Based High - Performance Computing on CUDA

In recent years, with the development of GPU, based on the general purpose computation on graphics processors has became a new field. Aiming at the processing of GPU, this paper provides the formal description for data parallel mode, a detailed description of the CUDA programming mode land the principle of optimization. It shows by the comparative experiment that CUDA owns strongly of the ability to the parallel processing and provides new methods and ideas to GPGPU.