This project aims to develop improved techniques for programming GPUs and to apply them in application domains such as financial engineering and audio signal processing.
Graphics processing units (GPUs) are ubiquitous in desktop platforms and serve to offload graphics processing requirements from the main processor and can also be used for general purpose computing. They have potentially much higher floating point performance than standard microprocessors as they have a large number of computational units in a single device. The main barrier for the adoption of computing using GPU technologies is that even for experts, they are difficult to program. In this work, languages, libraries and methodologies which make it easy to develop GPU-based applications in specialised domains will be studied. One area of interest is in financial option pricing where the parallel implementation of Monte Carlo and tree-based techniques can be studied. Another area, in collaboration with CARlab, will be audio signal processing applications.
The opportunity ID for this research opportunity is 1061