Parallelization of European Monte-Carlo options pricing on graphics processing units

Abstract

Using GPU computing for option pricing has been a critical problem for a long time, specifically, in Monte Carlo simulation which is the most widely used solution for option pricing problem. In general, option pricing must be performed in real time. Recent multi-core CPUs can provide a high computing power, but the option pricing on recent multi-core CPUs is far from responding in real time. The development of Graphic Processing Units promises a much higher computing power than multi-core CPUs for specialized problems. In this paper, we present methods to compress the data inputs for GPU in computing European options pricing which save large memory bandwidth and give the results in an acceptable time. The experimental result shows that the overall speedup is about 900X.