Earnshaw Medal
Earnshaw 2009 Nominees: Niall Donnelly, WIT
Real Time Modelling of Computational Fluid Dynamics using Compute Unified Device Architecture
With a shift from single to multi-core CPUs and now massively multi-core GPUs, programming paradigms must alter to harness this increased parallelism. Solutions to problems must now take into account the much increased parallelism of modern processors. One area that requires massive processing power is fluid dynamics and thus can benefit from this shift. The performance of previous distributed processing models (MPI) can now be enhanced using a new generation of massively parallel processors in graphics cards. The fluid dynamics problem that is implemented in this project deals with the flow of liquid past an object and the resulting evolution of vortices in its wake (known as a Karman vortex street). This is a standard fluid flow problem used to evaluate computational fluid dynamic (CFD) codes, but accurate generation of the resulting vortices requires significant computational effort. In this project, a parallel implementation on a CUDA based GPU is considered. As the Jacobi solver constitutes the majority of computational effort needed for the fluid solver, the effort in this project is directed at achieving a performance increase here. Three implementations of the solver are investigated, each increasing the performance of the solver.
^ To the top ^