Consumer Level Multi-GPU Systems Utilization, Efficiency, and Optimization

This paper focuses on the basic techniques that a common programmer can use to utilize multiple consumer level GPUs, commonly available to them at home or through purchase, to assist them in massively parallel computation in their work. It lays out an experimental framework to determine the best and easiest styles of programming for multiple GPUs to use, without actually using the more complex and available NVIDIA multi-GPU templates and techniques.