First principle simulation on many-core processors

Abstract

Many-core processors such as graphics processing units attract attention as the high-performance computational devices. Architectural development of computers over thirty years is briefly reviewed. Because of the breakdown of the CMOS scaling law, accumulation of simple CPU cores on a chip, called as many-core processors, becomes a rational way to achieve performance improvement. It is pointed out that not the absolute performance but the performance per watt matters near future. Next the most famous many-core processor, NVIDIA's FERMI, is introduced. It is a massively parallel processor,and is suitable for compute intensive, highly parallel, yet simple tasks. Several attempts to accelerate the computer simulations by using the GPU are reviewed, including the molecular dynamics simulations, first-principle calculation with plain wave basis set, and the quantum chemical calculations.