Abstract:Molecule dynamic (MD) simulation is used for simulation of atomic systems.It is a powerful tool for understanding relationships between microcosmic nature and macroscopic properties.In view of the problem of how to improve the performance of parallel MD simulations,in this paper taking the famous GROMACS software as an example,we study its implementation strategy on parallel computing performance,combining with the key principles of MD and the results of test cases,and we propose optimization strategies of performance under MPI and OpenMP parallel environment.It can provide theoretical and practical reference for optimizing performance of MD software.In addition,we also study that in the GPU heterogeneous parallel environment how to tune MPI/OpenMP and GPU to obtain the best performance.