Fugaku — The First Arm Supercomputer at No. 1

Fugaku (富岳) is a supercomputer that RIKEN and Fujitsu developed from 2014 as the successor to the K computer. On 22 June 2020 (Japan time), world rankings were announced at the online ISC2020 conference, and Fugaku took first place in all four of them: TOP500, HPCG, HPL-AI and Graph500. Its TOP500 LINPACK score was 415.5 petaflops, 2.8 times that of the second-place Summit in the United States.

At that point Fugaku was not finished and not yet open to users. It was completed and opened for shared use on 9 March 2021. The nine months between the day it became the world's fastest and the day it became available is the first thing to get straight if you are looking up dates.

The four rankings

The TOP500 ranks systems by their speed on LINPACK, which solves a large dense system of linear equations. The other three, HPCG (High Performance Conjugate Gradient), HPL-AI and Graph500, each measure performance on a different kind of computation. RIKEN presented first place in all four as evidence of Fugaku's all-round performance.

RIKEN said that taking first place in TOP500, HPCG and Graph500 at the same time was a world first; that is RIKEN's own claim. TOP500 noted that in single or further-reduced precision, often used in machine learning and AI, Fugaku's peak performance exceeded 1,000 petaflops (1 exaflops).

The first No. 1 on Arm

According to TOP500, Fugaku was the first No. 1 system powered by Arm processors. The K computer had used Fujitsu's SPARC64 VIIIfx; for Fugaku, Fujitsu moved to the Arm instruction set and designed its own 48-core A64FX. Arm is the architecture that came to dominate smartphones (see Arm architecture and ARM vs x86), and this was the first time it reached the top of the supercomputer rankings.

At the start of shared use, RIKEN President Hiroshi Matsumoto said he wanted to use Fugaku's world-leading performance and the versatility of the Arm architecture to widen the range of its work. Like the K computer, which took first place without GPUs or other accelerators, Fugaku reached No. 1 on CPUs alone.

COVID-19 work before completion

The last of Fugaku's racks were delivered in May 2020, and from April 2020, before completion, RIKEN offered its computing resources on a trial basis for COVID-19 countermeasures and other work. The best known was a project on predicting and countering droplet infection in indoor environments, led by Makoto Tsubokura, a team leader at the RIKEN Center for Computational Science and professor at Kobe University.

According to PC Watch's August 2020 report, the team used CUBE, a thermal-fluid analysis code implemented on Fugaku, to simulate how droplets and aerosols from coughing and speaking spread.

• A commercial non-woven mask completely blocked droplets larger than 50 microns; a polyester-like cloth mask blocked more than 90% and a cotton-like one about 80%
• Even with gaps at the cheeks and nose, a mask was reported to block 80% of droplets by volume
• Office partitions 1.2 m high were not enough to stop droplets, while partitions 1.4 m or higher tended to create poorly ventilated pockets

The results were released as videos and fed into public discussion of masks and ventilation. In its March 2021 announcement RIKEN wrote that this and related research were prompting changes in how people live.

Completion, shared use, and losing first place

When shared use began, RIKEN put Fugaku's application performance at up to 100 times that of the K computer and expected it to be used in drug discovery, weather, earthquakes and tsunamis, materials and more. Fugaku held the TOP500 lead for four lists in a row, through November 2021, and in June 2022 gave it up to Frontier at Oak Ridge National Laboratory in the US, which TOP500 called the first true exascale machine. In the June 2026 list, Fugaku is ninth at 442.01 petaflops.

The successor: FugakuNEXT

On 22 August 2025 RIKEN announced that it would design and develop FugakuNEXT (a development code name) with Fujitsu and NVIDIA. It will be the first Japanese flagship system to use GPUs as accelerators: NVIDIA leads the GPU platform design and Fujitsu the overall system and CPU design. RIKEN aims for operation around 2030 and more than five times Fugaku's hardware performance.

In the same announcement RIKEN looked back on the step from K to Fugaku: about 40 times more hardware performance and about three times more from improving the applications, which together exceeded the 100-fold target for some applications. It also noted that gains from semiconductor scaling are slowing, so large improvements from hardware alone are getting harder. The generation after Fugaku, which reached No. 1 on CPUs alone, will add GPUs to meet the demands of AI.

Other computing milestones of the era are on the hardware timeline.