Performance Evaluation of LAMMPS on Multi-core Systems

Abstract

With the widespread use of multi-core processors, specialized applications requiring parallel processing can be run on general desktops. In this study, we measure and analyze the parallel performance of LAMMPS, a classical well-known molecular dynamics code, on single multi-core systems. In order to check the parallel efficiency on single machines, the various types of simulations with LAMMPS were performed with MPI and OpenMP. Although LAMMPS was run on a single machine, MPI based LAMMPS showed higher performance, because LAMMPS has been designed based on MPI and only some part of LAMMPS was parallelized with OpenMP. The preliminary experiments also showed that the memory sub-system affects the performance of LAMMPS.