A data parallel approach to genetic programming using programmable graphics hardware

In recent years the computing power of graphics cards has increased significantly. Indeed, the growth in the computing power of these graphics cards is now several orders of magnitude greater than the growth in the power of computer processor units. Thus these graphics cards are now beginning to be used by the scientific community aslow cost, high performance computing platforms. Traditional genetic programming is a highly computer intensive algorithm but due to its parallel nature it can be distributed over multiple processors to increase the speed of the algorithm considerably. This is not applicable for single processor architectures but graphics cards provide a mechanism for developing a data parallel implementation of genetic programming. In this paper we will describe the technique of general purpose computing using graphics cards and how to extend this technique to genetic programming. We will demonstrate the improvement in the performance of genetic programming on single processor architectures which can be achieved by harnessing the computing power of these next generation graphics cards.