high performance computing on graphics processing units: hgpu.org

In the last few years, emergence of High-Performance Computing has largely influenced computer technology in the field of financial analytics, data mining, image/signal processing, simulations and modeling etc. Multi-threading, hyper-threading and other parallel programming technologies, multicore machines, clusters etc. have helped achieve high performance with high availability and high throughput. However, hybrid clusters have been gaining increased popularity these days due to the performance efficient, accelerated computation service they provide with the help of accelerators like GPUs (Graphics Processing Units), FPGAs (Field Programmable Gate Arrays), DSPs (Digital Signal Processors) etc. Amongst the accelerators, GPUs are most widely used for general purpose computations which require high speed and high performance as GPUs are highly parallel, multithreaded, manycore processors with tremendous computational power. In this paper, the hardware architecture of GPUs is discussed with light on how it is favorable to be used for general purpose computations. The programming models which are generally used these days for GPU computing and comparison of those models are discussed. Also a comparison of GPUs and CPUs in terms of both architecture and performance with the help of a benchmark is made and finally concluded with results.