high performance computing on graphics processing units: hgpu.org

We developed a lattice Boltzmann Solver, which can be used for the solution of low Reynolds number flow problems. Then, we modified it to run on Graphical Processing Unit using Compute Unified Device Architecture, which is a parallel computing platform and programming model created by NVIDIA. Comparison of the results that we obtained on Graphical Processing Unit and Central Processing Unit showed that the former was five times faster than the latter. In order to accelerate the solver further, we implemented multi-block capability, which allowed us to resolve the flow field around the object with a block of finer mesh while the rest of the computational domain consist of blocks of coarser mesh. We studied benchmarks problems of Couette flow, Poiseuille flow, lid driven cavity and flow around a cylinder confined in a channel. We obtained mesh independent results for the benchmark problems using multi and single relaxation time lattice Boltzmann solvers. As expected, multi relaxation time solver is more accurate and stable, but relatively slower.