high performance computing on graphics processing units: hgpu.org

Many computer graphics applications require high-intensity numerical simulation. The question arises whether such computations can be performed efficiently on the GPU, which has emerged as a full function streaming processor with high floating point performance. We show in this paper that this is indeed the case using two basic, broadly useful, computational kernels as examples. The first is a sparse matrix conjugate gradient solver and the second a regular-grid multigrid solver. Many realtime applications ranging from mesh smoothing and parameterization to fluid solvers and solid mechanics can greatly benefit from these as we demonstrate with a prototype implementation on NVIDIA