A Fast Double Precision CFD Code Using CUDA

We describe a second order double precision nite volume Boussinesq code designed to run on the CUDA architecture. We perform detailed validation of the code on a variety of Rayleigh-Benard convection problems and show second order convergence. We obtain matching results with a Fortran code running on an eight-core CPU. The CUDA-accelerated code performs approximately eight times faster than the Fortran code on identical problems. As a result, we are able to run a simulation with a grid of size 384 x 384 x 192 at 1.6 seconds per time step on a machine with a single GPU.