Smoothed particle hydrodynamics on GPU computing

Abstract

Smoothed Particle Hydrodynamics (SPH) is a powerful technique used to simulate complex free-surface flows. However one of the main drawbacks of this method is the expensive computational runtime and the large number of particles needed when 3D simulations are performed. High Performance Computing (HPC) therefore becomes essential to accelerate these codes and perform simulations. In this study, parallelization using Graphics Processing Units (GPU) is applied to the SPHysics code (www.sphysics.org) dedicated to free-surface flows with SPH. Simulations involving several million particles on a single GPU exhibit speedups of up to two orders of magnitude over the same calculations using CPU codes, while parallelization using MPI for multi-GPU leads to further acceleration. This cheap technology allows studying real-life engineering problems at reasonable computational runtimes.