COMPUTE ATLAS Supercomputer supply-chain graph
1153 systems 482 sites Sourced data

Parts/Accelerators/NVIDIA H100 SXM5 94GB

Accelerators · Hopper · 2023

NVIDIA H100 SXM5 94GB

94 GB HBM3 capacity variant of the H100 SXM5, used in TSUBAME4.0's node design.

Why this part

A higher-capacity SXM5 bin than the 80GB part this dataset otherwise tracks everywhere else (Eagle, Eos, MareNostrum 5 ACC, Jean Zay H100, Gefion, Njoerd); TSUBAME4.0's own hardware page and TOP500 independently agree on 94GB, so this is kept as its own part rather than folded into the 80GB row, the same way this dataset already splits the A100 SXM4 by its 40GB/64GB capacity variants.

Systems using this part

Systems
11
Rmax underneath
220.6 PFlop/s
Units recorded
3,060
7 of 11 edges disclose a quantity
System Role Country Units System Rmax Supplied by Rank
Capella
2024-11
Accelerator Germany - 24.06 PFlop/s NVIDIA #88 →
Virga
2024-06
Accelerator Australia - 14.94 PFlop/s NVIDIA #72 →
TSUBAME4.0
2024
Accelerator Japan - 39.62 PFlop/s NVIDIA #62 →
MUSICA
2024
Accelerator Austria 1,088 31.84 PFlop/s NVIDIA #50 →
Helma
2024
Accelerator Germany 384 32.22 PFlop/s NVIDIA #51 →
PITAGORA
2024
Accelerator Italy 672 27.90 PFlop/s NVIDIA #44 →
Kuma
2024
Accelerator Switzerland 336 12.31 PFlop/s NVIDIA #102 →
Cardinal
2024
Accelerator USA 128 5.04 PFlop/s NVIDIA #211 →
THE CRUST 2.5
2024
Accelerator Thailand - 13.85 PFlop/s NVIDIA #92 →
Lem
2024
Accelerator Poland 304 12.80 PFlop/s NVIDIA #80 →
Kraken-Fregata
2024
Accelerator Poland 148 5.98 PFlop/s NVIDIA #154 →

“Units recorded” sums only the edges where a public source states a quantity. It is a floor, not a total, and should never be read as installed base.

Further reading