Single particle cryo-EM emerges as a powerful and versatile method to characterize the structure and function of macromolecules, revealing the structural details of critical molecular machinery inside the cells. RELION is a widely used EM image processing software, and most of the recently published single particle cryo-EM structures were generated by using RELION. Due to the massive computational loads and the growing demands for processing much larger cryo-EM data sets, there is a pressing need to speed up image processing. Here we present GeRelion (https://github.com/gpu-pdl-nudt/GeRelion), an efficient parallel implementation of RELION on GPU system. In the performance tests using two cryo-EM data sets, GeRelion on 4 or 8 GPU cards outperformed RELION on 256 CPU cores, demonstrating dramatically improved speed and superb scalability. By greatly accelerating single particle cryo-EM structural analysis, GeRelion will facilitate both high resolution structure determination and dissection of mixed conformations of dynamic molecular machines.