Development of an overset grid computational fluid dynamics solver on graphical processing units

Abstract

General Purpose computation on Graphics Processing Units (GPGPUs) has gained popularity recently. Graphics Processing Units (GPUs) are being used for computationally intensive and data intensive problems to obtain orders of magnitude speed up in wide range of domains like molecular dynamics, bio-physics, geo-physics and CFD [1–4]. In this paper we discuss the development of a two-dimensional overset grid CFD solver on GPUs for moving body problems and demonstrate orders of magnitude speed-up on single GPU unit as compared to C/FORTRAN solver on a single CPU core. The two-dimensional overset grid CFD solver consists of three modules. A near-body solver module which solves the fluid conservation laws on structured and unstructured mesh systems, an off-body solver module which solves the fluid-conservation laws on an isotropic Cartesian mesh, and the domain connectivity module which manages the interaction between these two mesh systems. The GPU acceleration is extended to all the three modules. We expect this work to be the fore-runner for future development efforts for full three-dimensional Navier–Stokes solutions capable of executing in a heterogeneous parallel environment. Here “heterogeneous parallelism” refers to combination of Message Passing Interface (MPI) based communication for distributed memory systems, and large scale multi-threading using GPUs for shared memory systems.