American Options Pricing on Multi-core Graphic Cards

The aim of this paper is to explore the performances of graphics processing units (GPU) on the American options pricing problem using the Long staff and Schwartz method. This exploration includes a parallelization study of the different phases of American options pricing. We also give a comparison between CPU and GPU in pricing one-dimensional contracts. Finally, we investigate the running time of multidimensional contract pricing. We use NVIDIA Cg Toolkit for GPU programming and the comparison with CPU will be done against an open-source library implementation of the Long-staff and Schwartz algorithm.