high performance computing on graphics processing units: hgpu.org

With the decrease in cost of computing, and the increasingly friendly programming environments, the demand for computer generated models of real world problems has surged. Each generation of computer hardware becomes marginally faster than its predecessor, allowing for decreases in required computation time. However, the progression is slowing and will soon reach a barrier as lithography reaches its natural limits. General Purpose Graphics Processing Unit (GPGPU) programming, rather than traditional programming written for Central Processing Unit (CPU) architectures may be a viable way for computational scientists to continue to realize wall clock time reductions at a Moore’s Law pace. If a code can be modified to take advantage of the Single-Input-Multiple-Data (SIMD) architecture of Graphics Processing Units (GPUs), it may be possible to gain the functionality of hundreds or thousands of cores available on a GPU card. This paper details the investigation of a specific compressible flow simulation and its functionality in both CPU and GPU programming schemes. The flow is governed by the unsteady Euler flow equations and it is checked for validity against the known solution in all three directions. It is then run over varying grid sizes using both the CPU and GPU programming schemes to evaluate wall clock time reductions.