Parallel Computing on Clusters and Supercomputers

VMD supports large scale batch mode parallel analysis and visualization on clusters and supercomputers when it has been compiled with MPI support [14,35,26,40,41,36,37,38,42]. When running VMD on clusters and parallel computers it is possible to run one MPI rank per CPU core, or more likely, one MPI rank for several CPU cores, or one MPI rank for an entire compute node. If running more than one VMD instance per compute node, it is typically necessary to set environment variables to limit which CPU cores and/or GPUs each VMD instance attempts to use to prevent performance anomalies from arising due to resource contention [20].