Abstract: For the real-time SAR imaging systems, aiming at the problems of low real-time performance and high power consumption of traditional computing platforms, an implementation method was studied for embedded GPU. In order to make full use of the limited memory in the embedded GPU, a memory partitioning and reconfiguration scheme was proposed. The page-locked memory and zero-copy technology were used to realize the transmission-calculation parallelization. To achieve high real-time performance, large-scale parallelism was realized by using shared memory, registers, et. The result shows that sliding spotlight SAR imaging processing on the TX2 can only take 12.66 s time and consume 15 W power. This method is also applicable to other modes radar processing algorithms, and can provide reference for the future embedded real-time SAR imaging processing.