In this chapter, we want to demonstrate how NVIDIA’s compute unified device architecture platform can be used to reduce this CPU work and how it can be used to efficiently stream data between system memory and GPU memory. CUDA provides a straightforward way to address the GPU for general purpose computations, without the need to translate the problem in terms of shaders and textures. Although we implemented our system using CUDA, the theory will be equally applicable to upcoming vendor independent standards such as OpenCL or the DirectX Compute Shaders. The required hardware is getting cheaper and is no longer only available to the high-end market segment, making its use in commercial games attractive.