学際大規模情報基盤共同利用・共同研究拠点

採択課題 【詳細】

jh240029 Innovative Computational Science by Integration of Simulation/Data/Learning on Heterogeneous Supercomputers
課題代表者 中島研吾(東京大学情報基盤センター)
Kengo Nakajima (Information Technology Center, The University of Tokyo)
概要

In this project, we advance computational science by integrating "Simulation/Data/Learning (S+D+L)” using heterogeneous supercomputers like “Wisteria/BDEC-01 (U.Tokyo)” and “Flow (Nagoya U.)”, along with "mdx." In FY.2021 and FY.2022, our focus was earthquake simulation with real-time data assimilation. From FY.2023, we expand the "S+D+L" concept to other fields, initiating joint research and software development with international partners (Germany: JSC, FAU, France: CEA, Croatia: RBI). The project's leading PI and colleagues have been working on the innovative software platform “h3-Open-BDEC (https://h3-open-bdec.cc.u-tokyo.ac.jp/)” [3,4,5] since FY.2019 to facilitate "S+D+L" integration on heterogeneous systems like Wisteria/BDEC-01. Currently, Wisteria/BDEC-01 (“Odyssey” for Simulations with A64FX, “Aquarius” for Data/ML/AI with NVIDIA A100) operates as a full-scale platform for "S+D+L" integration with h3-Open-BDEC, gaining global attention. In FY.2023, we conducted research in atmospheric simulations, ported international partner applications to Wisteria/BDEC-01, and investigated "S+D+L" integration using h3-Open-BDEC. For FY.2024, we switch to the international project and focus on the fields of earth science, life science, and library/software/tool for integration of "S+D+L". JSC (Jülich Supercomputing Centre, Germany), one of the international partners of this project, also conducts research on heterogeneous computing based on the idea of ​​Modular Supercomputing Architecture. Some of the target applications in this project by JSC, such as TSMP and ChASE, are already feasible in heterogeneous environments.

報告書等 研究紹介ポスター / 最終報告書
関連Webページ
無断転載禁止