high performance computing on graphics processing units: hgpu.org

This paper is an introduction to general-purpose computing on graphics processing units. This involves taking advantage of the parallel processing power of modern graphics cards to do general purpose computation. The CUDA architecture used for general purpose computations on NVIDIA graphics cards is described, and important features affecting the run times of CUDA programs are discussed. Experimental results showing the potential for obtaining speedups by two or three orders of magnitude will be presented, showing that CUDA is a cost-effective way to make high-performance computing widely available to programmers and consumers. In colleges, graphics cards can be used to make hands-on experience in massively parallel processing an easily obtained component of courses or research programs.