high performance computing on graphics processing units: hgpu.org

This contribution concerns the verification and performance assessment of a hardware-accelerated parallel implementation of an algorithm for the semi-implicit finite difference method for solving the vertically integrated shallow water equations including a non-linear treatment of wetting and drying and conservative advection schemes. Instead of adapting an existing serial, OpenMP-, or MPI-parallelised code with all necessary compromises to be met, the selected approach is to write the code from scratch exposing the fine-grained parallelism of the scheme and execute the whole computational kernel of the code on a state-of-the-art streaming processor, i.e. a GPU. The reached speedups compared to a single CPU core are in the order of 20 or 30 for the double or single precision, respectively, which confirms the attractivity of the presently available advanced programming technologies for detailed, high-resolution river modelling applying commodity hardware.