high performance computing on graphics processing units: hgpu.org

The main objective of this project is to investigate the applications of heterogeneous acceleration to finite volume compressible flow solver for multiblock structured grids. Provided as Fortran source code, the ROTORMBMGS computational fluid dynamics program currently uses domain decomposition and message passing to distribute computation across multiple computers. Winning awards for scaling performance, there is little scope for improvement in parallel performance. However, there has been little investigation into accelerating the serial areas of the code. These serial areas provide an opportunity to take advantage of heterogeneous acceleration. OpenCL offers the ability to parallelise computation using both multi-core CPUs and the many-core architectures of graphics cards. Commonly offering a speed up of 10x or greater due to their large number of processing cores and high speed memory, graphics cards are becoming increasingly ubiquitous in supercomputers. This makes it critical to take advantage of these otherwise unutilised resources. Following acceleration of the program using OpenCL, the expectation is to see a significant improvement in performance which scales as more graphics cards are applied to the problem. Although low cost, consumer graphics cards can still offer significant acceleration, creating the possibility of running dramatically faster simulations on cheap consumer hardware outside of supercomputer clusters.