R-CCS Cafe 第209回 第2部

The research activities of the Imamura team are focused on the development and maintenance of high-performance numerical algorithms and software libraries for highly parallel computers such as Fugaku. The major three pillars of the research are 1) computational reproducibility in a broad sense, 2) portability in a broad sense, and 3) support for unexplored areas of numerical computation and hardware. For computational reproducibility, it is essential to have a hardware and software environment that guarantees computational accuracy, minimal precisions, and the ability to switch freely from low to high precisions arithmetic. We are collaborating with related research teams and external organizations. Some of the specific details were presented at the R-CCS Café last year. Portability means that the performance and functionality will continue to be guaranteed in the future, and an algorithmic approach and the use of hardware are required to achieve the equivalent or better functionality while maintaining higher performance. For example, to maintain EigenExa on the emerging systems is one of the motivated studies on this. Also, it is necessary to minimize the memory bottleneck of recent computers by data-compression using higher-dimensional arithmetic, modeling, and data quantization techniques and to reduce the problem representation itself to a lower order. The third, typical unexplored hardware is the quantum gate computer, which we recognize as a significant challenge for the future. In this R-CCS Café, I would like to report on the latest research results of 1) and 2) and possible technological developments that can be applied to other teams' activities.