Fine-Grained Parallel Computing Model Based on CUDA

As a bridge between natural model of an application and hardware model of computer,programming model is important in computer industry.Nevertheless,evolution of programming model could not suit already boosted GPU which have fine-grained parallel computing ability.The new launch of CUDA technology by NVIDIA GeForce 8 Graphics has freed GPGPU technology from the graphics fixed pipeline and high-level shader language,allowing the design and implementation of SIMD parallel algorithms on a much more simple way than previous method based on texture rendering.This paper focuses on fine-grained parallel computing model from character,composition and parallel architecture of CUDA.It illustrates that high performance computing based on GPGPU technology is an important direction of developments to meet the requirements of large scale computing.