Data processing is a time-consuming matter in the field of Synthetic Aperture Radar (SAR). In other ways, Graphic Processing Unit (GPU) have tremendous float-point computational horsepower and very high memory bandwidth, and the developing Compute Unified Device Architecture (CUDA) technology has enabled the GPU to be applied to the general purpose parallel computing. A new method of processing SAR data on GPU is presented in this paper. Compared with the nominal GPU based SAR processing method, number of data transfers between CPU/GPU are reduced from 4 to 1, and CPUs are exploited to cooperate with GPU synchronously. By the proposed method, data Processing is speeded up by 2.3 times, which is verified by the testing on the simulated SAR data.