Improvement of Fluid Simulation Runtime of Smoothed Particle Hydrodynamics by Using Graphics Processing Unit (GPU)

Abstract

This study concerns an implementation of smoothed particle hydrodynamics (SPH) fluid simulation on a graphics processing unit (GPU) using the Compute Unified Device Architecture’s (CUDA) parallel programming. A bookkeeping method for the neighbor search algorithm was incorporated to accelerate calculations. Based on sequence code profiling of the SPH method, particle interaction computation – which comprises the calculation of the continuity equation and the momentum conservation equation – consumes 95.2% of the calculation time. In this paper, an improvement of the calculation is proposed by calculating the particle interaction part on the GPU and by using a bookkeeping algorithm to restrict the calculation only to contributed particles. Three aspects are addressed in this paper: firstly, speed-up of the CUDA parallel programming computation as a function of the number of particles used in the simulation; secondly, the influence of double precision and single precision schemes on the computational acceleration; and thirdly, calculation accuracy with respect to the number of particles. Scott Russell’s wave generator was implemented for a 2D case and a 3D dam-break. The results show that the proposed method was succesfull in accelerating the SPH simulation on the GPU.