Overview

The Supercomputing Research Division is conducting research on hardware, software, algorithms, and applications of next-generation supercomputers.
The division also provides computational resources and related services as the core supercomputer center in Japan.

Innovative Supercomputing by Integration of
Simulation/Data/Learning (S+D+L)

Society 5.0 is a human-centered, safe, and secure society that closely integrates cyberspace and physical space to facilitate economic development and resolve social issues*1. While supercomputers have been mainly used for simulations in computational science, the integration of simulation with data analytics and machine learning will bring us closer to realizing Society 5.0.

Since around 2015, we have been engaged in the research and development for a supercomputer system that integrates simulations, data analytics, and machine learning (S+D+L). Known as the Big Data & Extreme Computing (BDEC) project, our first BDEC system, Wisteria/BDEC-01, commenced operations in May 2021. It comprises a simulation node group (Odyssey) with the same general-purpose CPU (A64FX) as the Supercomputer “Fugaku” and a data/learning node group (Aquarius) equipped with NVIDIA A100 Tensor Core GPUs. The system’s total peak performance is 33.1 petaflops. Wisteria/BDEC-01 is the world’s first supercomputer which combines different architectures for simulations and data analysis/machine learning. It features a shared file system with ample capacity and a fast SSD-based file system. Additionally, Aquarius is directly connected to an external network via a high-capacity communication line, enabling real-time observation data acquisition through SINET.

h3-Open-BDEC is an innovative software platform that integrates S+D+L by minimizing computation and energy consumption and maximizing the effectiveness of heterogeneous supercomputers such as Wisteria/BDEC-01. It is a collaborative development between computational science, computer science, numerical algorithms, data science, and machine learning. Wisteria/BDEC-01 and h3-Open-BDEC enable data assimilation by combining observation data analysis/utilization with simulations and optimizing model parameters. Leveraging machine learning can significantly reduce the need for multiple runs in real-world nonlinear simulations. Energy consumption is substantially reduced by merging high-performance, and highly-accurate/reliable computing methods with adaptive precision and accuracy verification. S+D+L integration spans diverse fields, including seismology, atmospheric science, materials science, and medicine, driving new scientific discoveries in these areas.

In January 2025, we plan to start operating the Miyabi (OFP-II) system at JCAHPC (Joint Center for Advanced HPC), which is jointly operated with the University of Tsukuba. Considering the global trend toward decarbonization in recent years and the soaring energy prices, ITC/UTokyo will be the first to introduce a system centered on highly energy-efficient GPUs. It is equipped with 1,120 units of the latest NVIDIA Grace-Hopper Superchip, and its total peak performance is 80.1 petaflops. At Miyabi, we aim to realize Society 5.0 by utilizing the knowledge and experiences from Wisteria/BDEC-01 regarding the integration of S+D+L on Miyabi using h3-Open-BDEC.

*1 Society 5.0: https://www8.cao.go.jp/cstp/english/society5_0/index.html