
Fugaku
A Japanese CPU-based supercomputer built by Fujitsu for RIKEN research work.
Fugaku is a large research supercomputer run by RIKEN at its Center for Computational Science in Kobe. Fujitsu built it as the follow-up to the K computer, with work starting in 2014 and the machine appearing in 2020. It led the TOP500 ranking from June 2020 to June 2022 and remains a major platform for Japanese science.
- 158,976 Fujitsu A64FX CPU nodes, no GPUs
- Held the TOP500 top spot from June 2020 to June 2022
- 442 PFLOPS on HPL after the November 2020 upgrade
- 2.0 EFLOPS on the mixed-precision HPL-AI benchmark
- Programme cost about US$1 billion (¥130 billion, 2018)
What it is
Fugaku is a scientific supercomputer operated by RIKEN at its computational science centre in Kobe, Japan. Fujitsu designed and supplied the hardware, which uses the company's own A64FX processor rather than graphics chips. Each of the 158,976 nodes carries one 48-plus-4-core CPU with 32 GiB of HBM2 memory. The machine serves open research work across many fields. Its measured speed reached 442 petaFLOPS on the standard TOP500 test after a late-2020 upgrade, and it hit 2.0 exaFLOPS on a mixed-precision AI benchmark. The whole programme cost about US$1 billion.
How it came about
Planning began in 2014 as a replacement for the K computer. RIKEN revealed the name Fugaku in May 2019, taken from another word for Mount Fuji, and showed a logo of the mountain that August. A prototype topped the Green500 efficiency list that November. Rack deliveries to the Kobe site ran from December 2019 to May 2020. The full system appeared in 2020 and entered routine service in 2021. It held the number one TOP500 spot from June 2020 until Frontier passed it in May 2022.
The compute
Every node holds a Fujitsu A64FX chip built on the ARM v8.2A design with Scalable Vector Extensions, aimed at roughly one exaFLOPS, about a hundred times the K computer. Nodes link through Fujitsu's Tofu interconnect D, a torus-style network. The OS is a lightweight multi-kernel setup called IHK/McKernel, running Linux beside McKernel so heavy simulations get a lean kernel while normal services stay on Linux. Storage has three levels: NVMe SSDs near the nodes, a 150 PB Lustre-based shared filesystem, and cloud storage services.
Where the GPUs are
The system sits at the RIKEN Center for Computational Science, known as R-CCS, in Kobe. Fujitsu shipped equipment racks there between December 2019 and May 2020. The programme is sponsored by Japan's Ministry of Education, Culture, Sports, Science and Technology (MEXT).
Customers and deals
Users come from across Japanese science. One early project studied how masks affect the spread of COVID-19. In May 2024 a Japanese team released Fugaku-LLM, a 13-billion-parameter language model with strong Japanese ability, trained on the machine after porting Megatron-DeepSpeed and tuning Transformer training for its CPU design.
Money and ownership
Nikkei reported in 2018 that the full programme would cost ¥130 billion, roughly US$1 billion at the time. RIKEN operates the system with backing from MEXT.
Its place in AI
Fugaku reached number one on the TOP500 in June 2020, the first ARM-based machine to do so, and stayed there until May 2022. In late 2020 it led four major rankings at once, including Graph500, HPL-AI and HPCG, a first for any system. By June 2026 it sat ninth on TOP500, just ahead of Alps.
Questions and criticism
The machine relies on CPUs alone, with no GPUs, so gains on AI-style benchmarks came from tuning rather than extra accelerators. Its position on TOP500 has slipped as newer systems arrived.
![Fugaku supercomputer at SC19. FUJITSU Supercomputer PRIMEHPC FX1000 Rack [1]](https://thumb.wikimedia.org/wikipedia/commons/thumb/c/c6/FugakuSupercomputerSC19.jpg/960px-FugakuSupercomputerSC19.jpg)
Questions people ask about Fugaku
What is Fugaku?
It is a research supercomputer run by RIKEN in Kobe, Japan. Fujitsu built it as the successor to the K computer. It uses Fujitsu's own A64FX ARM processors instead of graphics chips, and it supports scientific work across many fields.
How fast is it?
After a November 2020 upgrade it measured 442 petaFLOPS on the standard TOP500 test. On a mixed-precision AI benchmark it reached 2.0 exaFLOPS, the first result above one exaflop for any hardware at any precision.
Who paid for it?
Japan's Ministry of Education, Culture, Sports, Science and Technology sponsors the programme. Nikkei reported in 2018 that the total cost would be ¥130 billion, about US$1 billion.
What has it been used for?
Work includes COVID-19 mask research and the training of Fugaku-LLM, a 13-billion-parameter language model with strong Japanese skills, released in May 2024 by a Japanese team.
Good reading and sources
- Official websiteOfficial site
- Fugaku Remakes Exascale Computing In Its Own ImageThe Next Platform, Article
- Checking your browserAnandTech, Article
- お値段1300億円のポスト「京」、IT業界は今度こそ生かせるかNikkei, Article
- Fugaku (supercomputer) on WikipediaEncyclopedia
- More photos on Wikimedia CommonsPhotos
- Wikidata recordOpen data
The facts on this page come from the Wikipedia article "Fugaku (supercomputer)", written by its contributors and used under the CC BY-SA 4.0 licence, and from its Wikidata record (CC0). EveryNeocloud rewrote the text in its own words, with AI assistance, and compares every page with the article to make sure no passage is copied. It may be reused under the same CC BY-SA 4.0 licence with a link back.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.