Accelerating Unstructured Mesh Computational Fluid Dynamics on the NVidia Tesla GPU Architecture

Graham Markall
Imperial College London
Imperial College London, 2009


   title={Accelerating unstructured mesh computational fluid dynamics on the NVidia tesla GPU architecture},

   author={Markall, G. and Kelly, P.H.J.},


   institution={Technical report, Imperial College, London, UK}


Download Download (PDF)   View View   Source Source   Source codes Source codes




This report presents steps towards accelerating Fluidity, a general-purpose computational fluid dynamics package. One portion of the code, an iterative solver, is targeted for optimisation by using Graphics Processing Units (GPUs) to perform computations. A literature survey which examines the performance issues of iterative solvers and optimisations which may overcome these issues on classical and vector architectures is presented. Existing iterative solvers which use GPUs for computation are surveyed to identify optimisations which may accelerate our own solver implementation. The results of experimental investigations into improving an iterative solver which uses GPUs developed in a previous work is presented. It is shown that the speed of this solver compares favourably to the solver currently used in Fluidity, being able to solve large systems up to an order of magnitude faster. Numerical accuracy of the solver is shown to be limited, and its utilisation of the GPU solver shows room for improvement. Possible directions for further work which seeks to overcome these limitations is outlined.
No votes yet.
Please wait...

* * *

* * *

HGPU group © 2010-2020 hgpu.org

All rights belong to the respective authors

Contact us: