There is an increasing trend in modern space systems to move processing that until now was transmitted to ground for processing, on board the satellite. Synthetic Aperture Radar (SAR) is an example of such processing. However, such a computationally intensive task requires high performance hardware. In this paper we present the CPU and GPU acceleration of an on-board SAR processing application, part of ESA's open source benchmarking suite OBPMark. We benchmark several embedded multicore and GPU platforms which are promising candidates for future on-board systems. Our results show that both embedded multicores and especially GPUs can provide significant speedups in this type of processing, and can achieve performance levels similar to the ones of high performance ground stations.