high performance computing on graphics processing units: hgpu.org

High Performance Computing (HPC) is moving towards massive scales of parallelism. The changes in hardware towards large scale on chip parallelism requires the re-writing of existing solvers for various Computational Fluid Dynamics (CFD) problems. The aim of the project is to write and optimise novel solvers for various common CFD numerical problems that can take advantage of this increased parallelism on new platforms as they develop. Of special interest is the use of Graphics Processing Units (GPUs) which have only recently become available for use in HPC. GPU computing represents a significant change in the architectures used for scientific computing, and offers a factor of ten increase in peak performance/price ratio when compared with traditional HPC clusters. This report focusses on the optimisation of the geometric multigrid method on GPU and presents a literature review on the algebraic multigrid method and algorithms for parallelising this method. Work using a directive based GPU compiler to develop a GPU implementation of the Lattice Boltzmann is also shown.