A massively parallel molecular dynamics algorithm for the MasPar supercomputer

摘要

Massively parallel computers are emerging as a valuable tool for supercomputer applications. Their processing speed and memory size makes them ideal for solving large applications. An implementation of a molecular dynamics simulation using a neighbour list type algorithm is presented. By efficient use and understanding of the architecture, an extremely efficient neighbour list algorithm (without the need to store the list) has been developed. The large number of processors has allowed us to model large samples (up to one million atoms), reducing the artefacts which may be caused by having a small sample size. This implementation has provided performance results that surpass those of standard machines. The improvements are by factors of hundreds in terms of speed of calculation, and the sizes of the systems that can be modelled.