Fugaku

Fugaku
Name	Fugaku
Country	Japan
Developer	Riken and Fujitsu
Architecture	ARM-based A64FX many-core processor
Operators	Riken Center for Computational Science
Location	Kobe
Release date	2020
Peak performance	442 petaflops (HPL)
Memory	32 GB HBM2 per node
Interconnect	Tofu D
Os	Linux-based

Fugaku

Fugaku is a Japanese petascale-to-exascale supercomputer developed by Riken in collaboration with Fujitsu. Deployed at the Riken Center for Computational Science in Kobe, it was designed to advance research across domains including climate modeling, drug discovery, materials science, and disaster simulation. Fugaku topped multiple global rankings and informed national strategies in high-performance computing and computational science.

Overview

Fugaku was conceived as the successor to systems like K computer and intended to serve projects associated with institutions such as Japanese Cabinet Office, MEXT, and international collaborations with centers like NERSC and PRACE. Built around Fujitsu's A64FX processor, it emphasizes energy efficiency and scalability to meet workloads from initiatives led by groups such as Human Genome Project-era consortia, IPCC climate teams, and pharmaceutical partnerships including Takeda Pharmaceutical Company and Eli Lilly and Company. Its deployment supports research tied to programs from organizations like the European Centre for Medium-Range Weather Forecasts and agencies such as NASA and JAXA.

Design and Architecture

The hardware architecture combines Fujitsu-designed A64FX processors with the proprietary Tofu D interconnect and dense rack design inspired by predecessors like K computer. Each node integrates high-bandwidth memory patterned after HBM2 and coherence strategies analogous to designs in IBM Blue Gene and Cray XC series. The software stack leverages Linux ecosystems, compilers from Fujitsu and toolchains that interoperate with middleware used in projects like OpenMP, MPI, and runtimes common to CUDA-adjacent workflows. Fault tolerance and job scheduling incorporate influences from batch systems used at Oak Ridge National Laboratory and Lawrence Livermore National Laboratory, while the cooling and power delivery reflect industrial practices from firms such as Hitachi and Toshiba.

Performance and Benchmarks

Fugaku achieved top positions in rankings including TOP500, Green500, and HPCG through benchmarks run in coordination with researchers from Riken and international partners like Argonne National Laboratory. Its HPL results surpassed many contemporaries, reaching petaflop and near-exaflop-class performance reported alongside systems such as Summit (supercomputer), Sierra (supercomputer), Sunway TaihuLight, and Titan (supercomputer). Benchmarks for mixed-precision workloads referenced techniques from teams at NVIDIA and algorithmic developments linked to work from Stanford University and MIT. Energy-efficiency measurements compared Fugaku to installations at Lawrence Berkeley National Laboratory and academic clusters at University of Tokyo.

Applications and Use Cases

Fugaku has been applied across multidisciplinary projects, supporting simulations and data analysis for institutions like Osaka University, Kyoto University, University of Tokyo, and international collaborators such as Imperial College London and ETH Zurich. Use cases include pandemic-response modeling tied to efforts by World Health Organization researchers and pharmaceutical modeling in cooperation with companies including Pfizer and Roche. Climate and weather simulations have informed work by IPCC authors and operational centers like Meteorological Service of Canada and UK Met Office. Materials modeling, quantum simulations, and computational chemistry leveraged libraries and methods developed at Max Planck Society labs and Lawrence Livermore National Laboratory, while disaster resilience studies interfaced with agencies such as UNESCO and United Nations Office for Disaster Risk Reduction.

Development History

The project was announced following roadmap discussions involving organizations such as METI (Japan) and policy inputs from the Diet of Japan. Development teams at Riken and Fujitsu drew on lessons from the K computer program and industrial partnerships with semiconductor firms including Sony and fabrication collaborators akin to TSMC-style foundries. Software ecosystems were cultivated through collaborations with academic groups at University of Illinois Urbana-Champaign, University of California, Berkeley, and Tohoku University to ensure portability and adoption of standards popularized by communities around OpenMP and MPI. Funding and governance intersected with initiatives from entities like Japan Science and Technology Agency and research consortia modeled after EU Horizon projects.

Impact and Controversies

Fugaku influenced national strategy in supercomputing and spurred research outputs cited by agencies such as MEXT, Cabinet Office (Japan), and international science bodies including UNESCO and OECD. Its prominence stimulated competition with centers operating systems like Summit, Sierra, and Sunway TaihuLight, affecting procurement and R&D choices at corporations like Intel, AMD, and NVIDIA. Controversies included debates over procurement transparency debated in media outlets covering institutions such as NHK and policy forums involving Diet of Japan representatives, as well as discussions on export-control implications raised by stakeholders including Ministry of Economy, Trade and Industry (Japan) and international trade partners. Concerns about equitable access and research priority-setting were voiced by academics from University of Cambridge and Harvard University, and data governance issues paralleled dialogues at European Commission and U.S. National Science Foundation meetings.

Category:Supercomputers