The paper presents an implementation of the RELION image processing software that utilizes graphics processors (GPUs) to accelerate the most computationally intensive steps in cryo-EM structure determination. This includes image classification and high-resolution refinement, which have been accelerated by more than an order of magnitude on desktop hardware. The implementation reduces memory requirements to fit widely available consumer hardware and demonstrates that single precision arithmetic does not adversely affect the final results. These improvements enable high-resolution cryo-EM structure determination in a matter of days on a single workstation, rather than relying on large clusters. The authors also describe an improved algorithm for semi-automated particle selection, which further enhances the efficiency of the workflow. The GPU-enabled version of RELION-2 is designed to address current and future needs for large and expedient computations in cryo-EM, making it more accessible and efficient for researchers to perform their own reconstructions without access to supercomputers.The paper presents an implementation of the RELION image processing software that utilizes graphics processors (GPUs) to accelerate the most computationally intensive steps in cryo-EM structure determination. This includes image classification and high-resolution refinement, which have been accelerated by more than an order of magnitude on desktop hardware. The implementation reduces memory requirements to fit widely available consumer hardware and demonstrates that single precision arithmetic does not adversely affect the final results. These improvements enable high-resolution cryo-EM structure determination in a matter of days on a single workstation, rather than relying on large clusters. The authors also describe an improved algorithm for semi-automated particle selection, which further enhances the efficiency of the workflow. The GPU-enabled version of RELION-2 is designed to address current and future needs for large and expedient computations in cryo-EM, making it more accessible and efficient for researchers to perform their own reconstructions without access to supercomputers.