Molecular dynamics simulation of the supercooled Al melt on GPUs
Joint Institute for High Temperatures of RAS, Izhorskaya, 13, build. 2, Moscow 125412, Russia
Conference on Computational Physics, CCP 2010
@article{bystryimolecular,
title={Molecular dynamics simulation of the supercooled Al melt on GPUs},
author={Bystryi, R. and Kazennov, A. and Morozov, I. and Pisarev, V. and Stegailov, V.}
}
The method of molecular dynamics (MD) is widely used to study static and dynamic properties of the condensed matter [1]. In particular an approach to study the relaxation of metastable states is developed [2]. These states play essential role in the impulse loading processes such as shock compression, laser ablation, etc. Herewith we report on simulation technique and results for crystallization of the supercooled Al melt. As a phase transition (relaxation of the metastable state) in a single MD run is a random event one needs to perform the bunch of such runs starting from different microscopical configurations (which correspond to the same macroscopical state). Then the distribution of the metastable state lifetimes results in a physically meaningful value of the so called nucleation rate. Such code can be parallelized at two levels: force or domain decomposition for a single MD run and the parallel execution of a bunch of runs at different cluster (Grid) nodes. In this work we used the NVIDIA GPU for each MD run and executed the whole bunch on a multi-GPU workstation and a GPU cluster.
January 25, 2011 by hgpu
