NAG: Using GPU’s for Computational Finance
NAG Quant Event: The latest NVIDIA GPUs (graphics processors) have up to 240 cores on a single chip, and are fairly easily programmed using C/C++ with some extensions. In this talk I will discuss work with NAG to develop numerical routines for GPUs to generate pseudo-random and quasi-random numbers for use in Monte Carlo simulations, as well as my experiences in implementing simple 3D finite difference applications on GPUs. In both cases there is approximately a 10x improvement in speed, cost and energy efficiency compared to traditional servers with two quad-core CPUs. The potential of GPUs is underlined by Bloomberg’s recent announcement that they are using 192 GPUs for pricing calculations, instead of buying 1000 new servers.
Advertisement
