Promoted Fields Unit

Computer performance has been improved by increase of CPU (central processing unit) clock frequency based on miniaturization of semiconductor manufacturing process, however this reaches the upper limit caused by increase of heat generation, power consumption and leakage current since around 2005. To overcome these problems, in recent years, improvement of computer performance by parallel processing using multi-core processors which have multiple cores executing general operations at lower clock frequency has become a central issue. Since the year 2000, various processors have come on the market, for example multi-core CPUs, GPU (Graphics Processing Units) which enable fast numerical computations, and heterogeneous processors which combine an ordinary core for general oprerations with special cores for numerical operations. While such high performance processors based on parallel computing have come, ironically, development of numerical simulation software with high execution efficiency has been extremely difficult. In order to make full use of parallel hardware, we must select suitable calculation algorithms for the hardware architecture, and adjust programs to raise execution efficiency. This is called software tuning, which is an essential element for software development in high performance computing. There are technical difficulties as well as tremendous work in manual software tuning for extremely complicated hardware architecture in recent years. Furthermore, developed software must be rewritten and tuned in accordance with frequently updated hardware architecture to keep 4