Implementation of a GPU-based CFD code

Inspired by recent high performance computations via graphics processing units (GPUs), a co-located finite volume incompressible flow solver is implemented on a GeForce GTX480 chip with 480 processors. The CUDA programming standard is used on GPU and the CPU is employed through a C++ code. The lid driven cavity test case is used to measure the computational speed up due to the use of GPU as compared to a CPU-based solver. Two strategies have been examined to accelerate the numerical solution of the Navier-Stokes equations. In the first strategy, only the computations associated with the SIMPE algorithm have been transferred to GPU. In the second strategy all computations are carried out on GPU. Grid resolution and Reynolds number effects have also been investigated. Up to 40x speed up via GPU has been observed.