Multiparticle collision dynamics: GPU accelerated particle-based mesoscale hydrodynamic simulations

摘要

The Compute Unified Device Architecture (CUDA) programming language on a graphics processing unit (GPU) is exploited to develop a GPU-based simulation program for the multiparticle collision dynamics (MPC) approach, a particle-based mesoscale hydrodynamic simulation technique. The coarse-grained description of the fluid dynamics in terms of ballistic motion and local stochastic interactions of particles renders MPC inherently highly parallel. We achieve a 1-2 orders of magnitude performance gain over a comparable CPU-core version of the algorithm, depending on the implementation (single threaded or OpenMP). Various aspects of the implementation are discussed in the context of an optimized performance.