Efficient implementation of Constant False-Alarm Rate(CFAR) detector plays an important role in the development of signal processing system of new radar terminal. Under the Graphic Processing Unit(GPU) based software radar terminal architecture，this paper optimizes the CFAR algorithm implementation on GPU by using Compute Unified Device Architecture(CUDA) technology，which cuts data processing time tremendously compared to CPU implementation. The requirements of real-time for radar signal processing can then be satisfied，and the feasibility of the development of GPU based software radar terminal is verified.