News
DealsCorvex agrees USD 103 million take-or-pay deal to perform AI inference with Mid-Atlantic GPUsSovereignNeysa scales AI infrastructure in India with AMD Instinct MI350X GPUs on TytanInferenceMicrosoft Foundry adds Baseten, DeepSeek AI inference (MSFT:NASDAQ)ClustersDataOne and Rowan University sign engineering co-op program agreement for DataOne's Vineland AI factoryClustersLook inside Cal Poly's $3 million AI Factory (Video)DealsOpenAI Taps AMD Chips for 457 MW of AI Capacity in Six-Year TensorWave DealClustersCal Poly launches an AI data center on campus. Take a look insideInferenceArista takes aim at the next AI network battleground: distributed inferenceDealsOpenAI Signs 457 MW AI Data Center Deal as AMD Expands Its Challenge to NvidiaClustersBehind Dell AI Factory鈥檚 NVIDIA & AMD Enterprise ExpansionDealsCorvex agrees USD 103 million take-or-pay deal to perform AI inference with Mid-Atlantic GPUsSovereignNeysa scales AI infrastructure in India with AMD Instinct MI350X GPUs on TytanInferenceMicrosoft Foundry adds Baseten, DeepSeek AI inference (MSFT:NASDAQ)ClustersDataOne and Rowan University sign engineering co-op program agreement for DataOne's Vineland AI factoryClustersLook inside Cal Poly's $3 million AI Factory (Video)DealsOpenAI Taps AMD Chips for 457 MW of AI Capacity in Six-Year TensorWave DealClustersCal Poly launches an AI data center on campus. Take a look insideInferenceArista takes aim at the next AI network battleground: distributed inferenceDealsOpenAI Signs 457 MW AI Data Center Deal as AMD Expands Its Challenge to NvidiaClustersBehind Dell AI Factory鈥檚 NVIDIA & AMD Enterprise Expansion
EveryNeocloud
Racks (green mesh door) of interconnect switches of TSUBAME 3.0 supercomputer at Ookayama campus of the Tokyo Institute of Technology in Meguro-ku, Tokyo, Japan. Photographed on the open house for Koudaisai Tokyo Tech Fe

Tsubame

A long-running Japanese supercomputer series built for AI and scientific computing.

Photo: Kestrel, CC BY-SA 4.0, via Wikimedia Commons

Tsubame is a line of supercomputers run at the GSIC Center of the Institute of Science Tokyo, previously at Tokyo Institute of Technology. Designed by Satoshi Matsuoka, the series has been rebuilt several times since 2006 and now runs Tsubame 4.0, a large GPU system aimed at AI work.

Share on WhatsApp

  • First system started in 2006 at 85 TFLOPS, leading Japan and Asia
  • Tsubame 2.0 peaked at 2,288 TFLOPS with 4,200 Nvidia Tesla M2050 modules
  • Tsubame-KFC hit 4.5 gigaflops per watt using oil liquid cooling
  • Tsubame 3.0 topped the June 2017 Green500 at 14.110 GFLOPS/watts
  • Tsubame 4.0 runs 960 Nvidia H-100 GPUs and about 66.8 PFLOPS

What it is

Tsubame is a family of supercomputers housed at the GSIC Center within the Institute of Science Tokyo. The line began with a machine switched on in 2006 and has been replaced or upgraded repeatedly since. Satoshi Matsuoka designed the series. Each generation mixes general-purpose processors with accelerators, and the newest, Tsubame 4.0, went live in June 2024 with a strong focus on artificial intelligence work. The system serves research rather than commercial cloud rental.

How it came about

The first machine, put together by Sun Microsystems, started in 2006 and reached 85 TFLOPS, leading Japan and Asia at that moment. A 2008 upgrade added Nvidia Tesla hardware and lifted output to 170 TFLOPS. HP and NEC delivered a replacement in 2010, and a further refresh in 2014 swapped in Kepler K20x parts. An oil-cooled variant called Tsubame-KFC pushed efficiency records. SGI helped build a 2017 AI-focused system, and HP delivered the current machine in 2024.

The compute

Early nodes used dual-core AMD Opteron chips linked over InfiniBand, plus ClearSpeed accelerator cards. Later versions leaned heavily on Nvidia GPUs: Tesla T10, then M2050, then Kepler K20x, then P100. The 2024 machine pairs two 96-core AMD EPYC 9654 processors per node with four Nvidia H-100 GPUs and 768 GiB of DDR5 memory. Memory totals across older systems included 80.6 TB of DRAM plus GPU-side GDDR.

Where the GPUs are

All versions sit at the GSIC Center in Tokyo. The operator changed in October 2024 when the Institute of Science Tokyo was founded, taking over from Tokyo Institute of Technology. Tsubame 4.0 is the first in the series run by the new body.

Tsubame
A drawing made by EveryNeocloud from what this page says. It is a diagram, not a picture of the real provider.

Its place in AI

The series has repeatedly placed on the TOP500 ranking of the world's fastest machines. Tsubame 2.0 reached a high position on the November 2010 list and led Japan. Tsubame 3.0 appeared on the November 2017 list and topped the June 2017 Green500 efficiency table at 14.110 GFLOPS per watt. Tsubame 4.0 entered the November 2024 TOP500 at 36th worldwide and 5th within Japan.

Questions and criticism

Power use has been a design concern. The Tsubame-KFC variant used oil-based liquid cooling to cut consumption and reached 4.5 gigaflops per watt, described at the time as the best efficiency figure worldwide.

Questions people ask about Tsubame

What is Tsubame?

It is a series of supercomputers operated at the GSIC Center of the Institute of Science Tokyo. The line started in 2006 and has been rebuilt several times, with the current version running since June 2024 and aimed at AI work.

Who builds these machines?

Different vendors handled each generation. Sun Microsystems built the first, HP and NEC delivered the 2010 replacement, SGI worked on the 2017 system, and HP produced Tsubame 4.0. Satoshi Matsuoka designed the series.

What hardware does the newest system use?

Tsubame 4.0 has 240 nodes. Each node carries two 96-core AMD EPYC 9654 processors, 768 GiB of DDR5-4800 memory and four Nvidia H-100 GPUs, giving roughly 66.8 PFLOPS in double precision.

How has it ranked internationally?

Tsubame 2.0 reached a high mark on the November 2010 TOP500 and led Japan. Tsubame 3.0 placed 13th in November 2017 and led the June 2017 Green500. Tsubame 4.0 ranked 36rd worldwide on the November 2024 list.

Good reading and sources

Where this page comes from

The facts on this page come from the Wikipedia article "Tsubame (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.

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 provider of the day, new pages on EveryNeocloud, and the conferences and anniversaries of the week ahead.