With the characteristics of large data volume, high algorithm complexity and large computational complexity, Synthetic Aperture Radar (SAR) technology which makes the signal processing system have to be improved continuously in the aspects of real-time, storage capacity, data throughput and computing capability. As a kind of multi-core architecture, Graphics Processing Unit (GPU) take the advantages of powerful computing capability and efficient storage bandwidth to meet the urgent need in scalability, computing capability and storage bandwidth for large-scale data parallel applications. In this paper, the first thing is to evaluate the FFT performance of the NVIDIA Tesla M6 GPU, which achieves an average 41x speedup ratio compared to TI’s TMS320C6678 DSP. Then, the RD (Range Doppler) algorithm which is the most classical SAR imaging algorithm is implemented on the platform of CPU + GPU using CUDA language, and execution time of the SAR algorithm for 4 K × 8 K point is shortened by 1.18 s and the result shows that GPU achieve 1.9x the performance improvement compared to DSP C6678 on RD-SAR algorithm.