A GPU-enabled Finite Volume solver for global magnetospheric simulations on unstructured grids

摘要

This paper describes an ideal Magnetohydrodynamics (MHD) solver for global magnetospheric simulations based on a B_1+B_0 splitting approach, which has been implemented within the COOLFluiD platform and adapted to run on modern heterogeneous architectures featuring General Purpose Graphical Processing Units (GPGPUs). The code is based on a state-of-the-art Finite Volume discretization for unstructured grids and either explicit or implicit time integration, suitable for both steady and time accurate problems. Innovative object-oriented design and coding techniques mixing C++ and CUDA are discussed. Performance results of the modified code on single and multiple processors are presented and compared with those provided by the original solver.