Utilizing Graphics Processing Units (GPUs) for Numerical Computations

Abstract

Initially designed for graphics processing, GPUs (graphics processing unit) only contained fixed rendering functions. As they are parallel processors with high computational power for arithmetic calculations, GPUs have been evolving rapidly with the inclusion of programmable pipelines. The trend is for the GPU to become a generic processor (GPGPU). Furthermore, the integration between GPU and CPU cores within a single chip also tends to become a standard among processors. GPUs are also becoming the target of massively parallel systems such as supercomputers. Today, the fastest supercomputer in the world has thousands of GPUs, which has significantly increased its performance compared to its predecessors.