high performance computing on graphics processing units: hgpu.org

We present CUDACLAW, a data-parallel solution framework for 2D and 3D hyperbolic partial differential equation (PDE) systems. CUDACLAW is a finite volume method based on time adaptive point-wise Riemann problem solvers, and can handle linear and nonlinear problems. The framework is tailored for the GPU architecture, optimized to take advantage of the powerful computational potential, high memory bandwidth and fast on-chip memories. Our solution approach minimizes total memory usage allowing large problems to run on commodity hardware. From the user’s point of view, the framework is simple and its use does not require any knowledge of the underlying data structure, nor experience in CUDA programming. With the power of the graphics processor, the solution can be viewed in real time. We demonstrate the power of CUDACLAW on an acoustics wave equation solution in 2D and 3D. Sustained performance of more than a 150 GFlops-s is achieved on the Tesla C2050 Nvidia GPU.