A simulation of the Milky Way galaxy, utilizing exascale supercomputers and AI, advances to Finals for 2026 ACM Gordon Bell Prize

The Gordon Bell Prize, a leading award in the field of high-performance computing (HPC), is presented to research groups that have achieved particularly outstanding results by applying high-performance computing to science and technology problems. For this study, they used three different exascale supercomputers, each featuring distinct GPU architectures—Aurora at Argonne National Laboratory (ANL) in the U.S., which is equipped with Intel GPUs; Frontier at Oak Ridge National Laboratory (ORNL) in the U.S., which is equipped with AMD GPUs; and JUPITER at the Jülich Research Center (FZJ) in Germany, equipped with NVIDIA GPUs—to run ultra-high-resolution simulations of the Milky Way galaxy. On Aurora, 4.8 trillion particles were simulated using 10,040 nodes, processing up to 144 billion particles per second and achieving a computational performance of 976 PFLOP/s in gravitational calculations. Meanwhile, Frontier carried out a simulation using 5.4 trillion particles—the largest in this study. Additionally, on JUPITER, the authors processed 135 billion particles per second using 5,184 nodes, achieving high throughput approaching that of Aurora with approximately half the number of nodes. In addition, they introduced a surrogate model that uses AI to predict gas expansion following a supernova explosion, thereby streamlining the simulations by reducing the increase in local computational cost. As a result, they successfully ran simulations that reproduced every single star in the Milky Way on all three exascale supercomputers, despite their differing GPU vendors and computing architectures.

This work paves the way for future ultra-high-resolution simulations that utilize a variety of exascale computers to trace the formation and evolution of the Milky Way on the scale of individual stars.

The winners of the Gordon Bell Prize are scheduled to be announced at SC26 (The International Conference for High Performance Computing, Networking, Storage, and Analysis), an international conference to be held in Chicago, Illinois, from November 15 to 20.

International Collaborative Research Group

RIKEN Center for Interdisciplinary Theoretical and Mathematical Sciences, Division of Fundamental Mathematical Science
Special Postdoctoral Researcher Keiya Hirashima

Graduate School of Science, The University of Tokyo
Professor Michiko Fujii
Project Researcher Naoto Harada
Assistant Professor Kana Moriwaki

Information Technology Center, The University of Tokyo
Associate Professor Yohei Miki

Graduate School of Science, Kobe University
Associate Professor Takayuki Saitoh
Project Professor Junichiro Makino

Preferred Networks, Inc.
Engineer Kentaro Nomura

Department of Information Engineering, National Institute of Technology, Matsue College
Associate Professor Masaki Iwasawa

Scientific Computing Core, Flatiron Institute
HPC Optimization Engineer Dennis Bollweg

Center for Computational Science, University of Tsukuba
Associate Professor Kohji Yoshikawa

Institute of Space Science, University of Barcelona (Spain)
Postdoctoral Researcher Tetsuro Asano

Faculty of Community Service and Science, Tohoku University of Community Service and Science
Lecturer Yutaka Hirai

Graduate School of Science, Hokkaido University
Professor Takashi Okamoto

NVIDIA
Engineer Matthew Bettencourt
Engineer Laura Morgenstern

Jülich Supercomputing Centre
Mathis Bode

Project details

Visualization of the simulation

Comment from the Finalist

Keiya Hirashima, Special Postdoctoral Researcher, Division of Fundamental Mathematical Science, iTHEMS

Upon learning the group has been selected, Dr. Hirashima said, “I am extremely pleased and honored that the research I have been working on since entering the University of Tokyo’s graduate school in 2020 has been selected as a finalist for the Gordon Bell Prize. The galactic simulation code that forms the foundation of this research was developed through many years of collaborative research using multiple supercomputers including Supercomputer Fugaku. I began my research based on this code in 2020, with the goal of directly tracking the impact of individual stellar physics—including supernova explosions—on the evolution of entire galaxies. In 2023, we released an AI model that accelerates supernova explosion simulations, and in 2025, achieved the world’s first galaxy simulation combined with AI, while also realizing high parallelization performance on Fugaku. In this latest research, we have made further improvements, enabling us to run galaxy simulations on a larger scale and with finer detail than ever before on exascale supercomputers around the world equipped with different GPUs. This allows us to bridge an extremely wide range of scales—from individual stars to the entire galaxy—and directly track the process by which stellar activity shapes the evolution of the Milky Way, our home galaxy. Through this process, we have carried out collaborative research with researchers from around the world, including in New York, where I am currently based, and our research group has become increasingly international. I am very pleased that these results have been recognized, and am grateful to the many collaborators who have supported our research and advanced our development efforts.”

Supplementary Explanation

Exascale supercomputer

A supercomputer capable of performing more than one quintillion (10¹⁸) floating-point operations per second.

Aurora

A supercomputer installed at Argonne National Laboratory of the U.S. Department of Energy, equipped with Intel CPUs and GPUs. Its computing performance ranks 4th in the world (TOP500; as of September 2026).

Frontier

A supercomputer installed at Oak Ridge National Laboratory of the U.S. Department of Energy, equipped with AMD CPUs and GPUs. In 2022, it became the first in the world to achieve "exascale" computing performance. Its computing performance ranks 3rd in the world (TOP500; as of September 2026).

JUPITER

Europe's first "exascale" supercomputer, installed at the Jülich Research Center in Germany. Its computing performance ranks 5th in the world (TOP500; as of September 2026).

Supernova explosion

A spectacular explosion that occurs when a massive star ends its life. The explosion releases heavy elements created inside the star into space, where they become the raw materials for new stars and planets. The elements produced in this way drive the evolution of galaxies and support the formation of diverse celestial objects.

Interstellar medium

A general term for the gas and dust that fill the space between stars within a galaxy. Even space that appears empty is filled with tenuous gas, which exists in diverse states, from cold, dense regions to hot, diffuse ones. Stars are born when this interstellar medium gathers, and it is stirred up again by supernova explosions and radiation from stars. This cycle shapes the formation of new stars and the evolution of entire galaxies.