The spherical harmonics model to describe geopotentials relies on recursive algorithms,whose computational efficiency degrades significantly if the order of the expansion into spherical harmonics increases.In order to enhance the computational efficiency for high-fidelity geopotentials,a Global Point Mascon model and a fast algorithm using GPU parallel computing are presented in this paper.The GPU parallel computing is implemented with CUDA (Compute Unified Device Architecture),enabling fast computation of high-fidelity geopotentials.The spherical harmonics and the Global Point Mascon models are evaluated and compared by conducting numerical integration of space trajectories,which shows that the latter model with parallel computing leads to one order of magnitude speedup without degradation of numerical integration accuracy.