Abstract:The focus of the current paper is the development of a finite volume computational fluid dynamics solver on the Graphical Processing Unites (GPUs) using CUDA Fortran for turbulence simulations. The solver is implemented with an AUSM scheme for the spatial discretization and the SST kappa-omega model for turbulence model. Several test cases, such as 2D NASA 0012 airfoil, a supersonic mixing layer, and hypersonic inlet flow are chosen to demonstrate the acceleration performance of GPU on multi-million grid cells. Results show that the computational expense can be reduced by 20-32 times when using a NVIDIA Tesla K20c GPU as compared to a single core of an Intel Xeon 2670 CPU.

A Cfd Solver On Graphical Processing Unites For Turbulence Simulations