Open Access
Subscription Access
A Fast Double Precision CFD Code Using CUDA
Abstract
We describe a second order double precision finite volume Boussinesq code designed to run on the CUDA architecture. We perform detailed validation of the code on a variety of Rayleigh-B´enard convection problems and show second order convergence. We obtain matching results with a Fortran code running on an eightcore 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 3842 × 192 at 1.6 seconds per time step on a machine with a single GPU.