News
IndustryTSMC’s grip on advanced chipmaking will face competitionIndustryCadence joins European Chip Design Platform to support semiconductor startups and SMEsMaking chipsMarc Hijink on ASML’s Culture, Supply Chain, and GeopoliticsProcessorsIntel's post-Nova Lake CPUs take shape as Razor Lake desktop and laptop variants surfaceAI chipsHow Google’s New AI Chip Software Helps It Battle NvidiaIndiaMore IC Packaging Plants Emerge In IndiaIndustryUS Tightens China AI Chip Access as Nvidia, TSMC and Semiconductor Stocks SlideDesignAxelera AI moves into HPC with TitaniaMaking chipsChinese envoy says chip equipment export curbs threaten EU-China trade thawIndiaSEMICON India 2026 Day 2 Sees 25 MoUs, Strategic Collaborations Across Semiconductor Value ChainIndustryTSMC’s grip on advanced chipmaking will face competitionIndustryCadence joins European Chip Design Platform to support semiconductor startups and SMEsMaking chipsMarc Hijink on ASML’s Culture, Supply Chain, and GeopoliticsProcessorsIntel's post-Nova Lake CPUs take shape as Razor Lake desktop and laptop variants surfaceAI chipsHow Google’s New AI Chip Software Helps It Battle NvidiaIndiaMore IC Packaging Plants Emerge In IndiaIndustryUS Tightens China AI Chip Access as Nvidia, TSMC and Semiconductor Stocks SlideDesignAxelera AI moves into HPC with TitaniaMaking chipsChinese envoy says chip equipment export curbs threaten EU-China trade thawIndiaSEMICON India 2026 Day 2 Sees 25 MoUs, Strategic Collaborations Across Semiconductor Value Chain
EveryChip
Fujitsu A64FX

Fujitsu A64FX

48-core ARM supercomputer CPU with 512-bit SVE and HBM2, built for Fugaku

The A64FX is a 64-bit ARM microprocessor designed by Fujitsu and manufactured by TSMC on a 7 nm process, entering production in 2019. It replaced the SPARC64 XIfx as Fujitsu's supercomputer chip and powered Fugaku, the world's fastest TOP500 machine from June 2020 until 2022.

Share on WhatsApp

  • 48 compute cores per processor, arranged as four NUMA nodes of 12
  • 512-bit Scalable Vector Extension units on the ARMv8.2-A instruction set
  • 32 GB of HBM2 memory with 1 TB/s bandwidth
  • About 8.786 billion transistors, built on a TSMC 7 nm process
  • Powered Fugaku, the top TOP500 supercomputer from June 2020 to 2022

What it is

The A64FX is a 64-bit ARM microprocessor built by Fujitsu for supercomputers. It implements the ARMv8.2-A instruction set together with 512-bit Scalable Vector Extension (SVE) SIMD units and meets Server Base System Architecture level 3. Fujitsu designed it to take over from the SPARC64 XIfx in its high performance computing line. The chip is best known for powering Fugaku, which topped the TOP500 list of the world's fastest supercomputers from June 2020 until it was overtaken by Frontier in 2022. Fujitsu also sells smaller machines built around the processor.

How it came about

Fujitsu developed the A64FX specifically for the Fugaku supercomputer. It entered production in 2019 as the successor to the SPARC64 XIfx. In June 2020 a Fugaku system using these processors reached 442 petaFLOPS and took the top spot in the TOP500 rankings. Fujitsu has said it intends to sell smaller machines with A64FX chips, and Cray is developing supercomputers that use them.

Inside the chip

Each processor is split into four NUMA nodes, meaning blocks of memory and cores that behave as separate units. Every node holds 12 compute cores, giving 48 cores in total, plus assistant cores that handle non-computational work and its own level 2 cache and HBM2 memory. The SVE vector units take four operands rather than the usual three; the extra one is an instruction prefix, and a MOVPRFX instruction is packed with the following operation in the pipeline. Fujitsu claims over 90% execution efficiency on matrix multiply and integer dot product work. Around 128,400 error checkers support reliability features.

How it was made

TSMC manufactures the A64FX on a 7 nm process. The die carries roughly 8.786 billion transistors. It is paired with 32 gigabytes of HBM2 memory offering 1 TB per second of bandwidth, and includes 16 PCIe 3.0 lanes for NVMe solid state storage. A built-in TofuD fabric controller provides 10 ports, implemented as 20 lanes running at 28 Gbit/s, to link nodes into a cluster.

Where it went

The main deployment is the Fugaku supercomputer, which led the TOP500 rankings between June 2020 and 2022. Fujitsu offers the PRIMEHPC FX700 server, a two-node A64FX machine. Cray is building systems with the chip, including Isambard 2 in the United Kingdom for a consortium led by the University of Bristol that also includes the Met Office. Ookami, an NSF-backed open testbed run by Stony Brook University and the University at Buffalo, gives researchers access to A64FX hardware.

Versions and family

The A64FX replaces the SPARC64 XIfx in Fujitsu's supercomputer range. Fujitsu plans lower specification versions with fewer assistant cores. Isambard 2 upgrades the earlier Isambard machine, which used the Marvell ThunderX2 ARM processor. Other ARM chips aimed at similar work include the Huawei-owned HiSilicon Kunpeng 920, sold only in China.

Why it matters

Fugaku, built from A64FX processors, held first place in the TOP500 list of the world's fastest supercomputers from June 2020 until Frontier passed it in 2022. The machine reached 442 petaFLOPS in June 2020. The chip showed that an ARM design could lead the supercomputer rankings, and it spread beyond Fugaku into UK and US research systems.

Problems and reception

Anandtech reported in June 2020 that a PRIMEHPC FX700 server with two A64FX nodes cost ¥4,155,330, around US$39,000 at the time. Fujitsu claims high execution efficiency on dense matrix and integer dot product code, and reliability features backed by roughly 128,400 error checkers.

Questions people ask about Fujitsu A64FX

What kind of chip is the A64FX?

It is a 64-bit ARM microprocessor designed by Fujitsu for supercomputers. It runs the ARMv8.2-A instruction set with 512-bit SVE vector units, and it was made to replace the SPARC64 XIfx in Fujitsu's high performance computing line.

How is the processor organised?

Each chip has four NUMA nodes. Every node contains 12 compute cores, so 48 cores in total, plus assistant cores for non-computational tasks. Each node also has its own level 2 cache and its own HBM2 memory.

What memory does it use?

The A64FX is paired with 32 gigabytes of HBM2 memory, giving a bandwidth of 1 TB per second. It also has 16 PCIe 3.0 lanes for NVMe solid state storage and a TofuD fabric controller for connecting nodes.

Where has it been used?

Its main home is the Fugaku supercomputer. Cray is building systems with it, including Isambard 2 in the UK, and the Ookami testbed at Stony Brook University and the University at Buffalo offers researchers access to the processors.

Who manufactures it?

Fujitsu designed the A64FX and TSMC manufactures it on a 7 nm process. The chip carries roughly 8.786 billion transistors and entered production in 2019.

Good reading and sources

Where this page comes from

The facts on this page come from the Wikipedia article "Fujitsu A64FX", written by its contributors and used under the CC BY-SA 4.0 licence, and from its Wikidata record (CC0). EveryChip 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.

Photo credits on this page

The small pictures above come from Wikimedia Commons under free licences. Some are cropped to fit.

The weekly digest

Every Monday: the chip of the day, new pages on EveryChip, and the conferences and anniversaries of the week ahead.