Computational finance acceleration using CUDA on graphics processing units

The computation of fair prices for options has become an increasingly intrinsic aspect of finance today. In an era of supercomputing, there is a demand to perform calculations quicker and achieve accurate results to complex problems like never envisaged previously. GPUs and FPGAs have dawned as hybrid elements to accelerate risk analysis and have succeeded in speeding up models and achieving a far greater throughput. Such is the power of this new technology that traders can now evaluate the risk on their books in a matter of seconds as opposed to overnight in the previous era of computing. This enables them to respond instantaneously to changes in the market and make quicker decisions yielding higher profi ts. The objective of this project was to understand the Black-Scholes model and the relevance of the GPU in evaluating options prices. This would not only help comprehend the architecture of the modern GPU but also lead to the development of a model finely tuned and optimised for the highest performance. Thread structure optimisations were performed and several key features of the Tesla GPU were used to produce an advanced and scalable optimum solution. The model developed exhibited an 1800x speed up over the CPU and it was shown that an increase in accuracy of the options prices invariably led to a large decrease in computational speed up. The response of the model under several conditions and changing parameters led to a deeper understanding of the GPU, its bene fits as a computational workhorse and certain drawbacks like kernel and memory overhead. It is recommended for future studies that several option pricing models be compared against one another to gauge their eff ective speed up and relative accuracy. This can provide financial fi rms the competitive advantage they need in eff ciently predicting market movement.Bachelor of Engineering (Computer Engineering