high performance computing on graphics processing units: hgpu.org

The processing power of graphics hardware has increased tremendously in the last several years and they are therefore used more and more outside of their intended domain of graphics rendering. This thesis describes the implementation and results of a fluid simulator, using the constraint fluid method, which harnesses the processing power of modern GPUs, in particular NVIDIA’s CUDA platform. As demonstrated in this thesis, particle systems with hundreds of thousands of particles can be simulated and visualized at interactive rates and systems containing up to a million particles can be run at a few frames per second. The biggest performance bottleneck is currently in the solver, in particular the lack of a working preconditioned Conjugate Gradient implementation.