Parts/Accelerators/NVIDIA H100 SXM5
Accelerators · Hopper · 2022
NVIDIA H100 SXM5
80 GB HBM3 SXM module. The unit of account for essentially every commercial AI training cluster built 2023-2024.
Why this part
The most commercially important processor of its decade, and the point at which supply rather than design became the constraint: for roughly two years the question was not whether you wanted H100s but whether you could get them. It added FP8 and a transformer engine that switches precision per layer, both aimed squarely at large language models. Its 700W envelope also settled an infrastructure argument, because eight of these in a chassis cannot be air cooled in any practical rack, and Meta's two published clusters, one on RoCE and one on InfiniBand, are the clearest public evidence of what an operator has to decide around it.
Systems using this part
- Systems
- 81
- Rmax underneath
- 2.21 EFlop/s
- Units recorded
- 247,580
43 of 81 edges disclose a quantity
| System | Role | Country | Units | System Rmax | Supplied by | Rank |
|---|---|---|---|---|---|---|
| Shakti Cloud 2025-11 | Accelerator | India | - | 84.31 PFlop/s | NVIDIA | #32 → |
| Otus 2025-11 | Accelerator | Germany | 108 | 4.66 PFlop/s | NVIDIA | #5 → |
| Harpia 2025-10 | Accelerator | Brazil | - | 75.20 PFlop/s | NVIDIA | #37 → |
| Trillium 2025-08-07 | Accelerator | Canada | 252 | 9.49 PFlop/s | NVIDIA | #175 → |
| Rorqual 2025-07 | Accelerator | Canada | 324 | 6.97 PFlop/s | NVIDIA | #192 → |
| ABCI-Q 2025-06 | Accelerator | Japan | - | 74.58 PFlop/s | NVIDIA | #38 → |
| SSC-24 2025-06 | Accelerator | South Korea | - | 106.2 PFlop/s | NVIDIA | #25 → |
| Space HPC 2025 | Accelerator | Italy | 108 | 4.47 PFlop/s | NVIDIA | #261 → |
| TX-GAIN 2025 | Accelerator | USA | - | 13.39 PFlop/s | NVIDIA | #114 → |
| Bank of America HPC-AI 2025 | Accelerator | USA | - | 21.46 PFlop/s | NVIDIA | #76 → |
| Matrix 2025 | Accelerator | USA | 112 | 3.11 PFlop/s | NVIDIA | #406 → |
| TII Oryx 2025 | Accelerator | United Arab Emirates | - | 19.48 PFlop/s | NVIDIA | #90 → |
| SuperPOD 2024-11 | Accelerator | United Arab Emirates | - | 55.81 PFlop/s | NVIDIA | #46 → |
| FPT AI Factory Vietnam 2024-11 | Accelerator | Vietnam | - | 46.65 PFlop/s | NVIDIA | #38 → |
| Sesterce Marseille SuperComputer 2024-11 | Accelerator | France | - | - | NVIDIA | - |
| Gefion 2024-10 | Accelerator | Denmark | - | 66.59 PFlop/s | NVIDIA | #43 → |
| CHIE-2 2024-10 | Accelerator | Japan | - | 89.78 PFlop/s | NVIDIA | #30 → |
| Tesla Cortex 2024-10 | Accelerator | USA | 50,000 | - | NVIDIA | - |
| ASPIRE 2A+ 2024-10 | Accelerator | Singapore | 320 | 14.20 PFlop/s | NVIDIA | #90 → |
| Arka AI/ML 2024-09 | Accelerator | India | - | 2.70 PFlop/s | NVIDIA | #400 → |
| NeXXt AI Factory 2024-07-09 | Accelerator | Italy | 248 | - | NVIDIA | - |
| xAI Colossus 2024-07 | Accelerator | USA | 100,000 | - | NVIDIA | - |
| Jean Zay H100 2024-06 | Accelerator | France | - | 52.18 PFlop/s | NVIDIA | #49 → |
| BioHive-2 2024-05 | Accelerator | USA | 504 | 23.32 PFlop/s | NVIDIA | #35 → |
| IBM Blue Vela 2024-04 | Accelerator | USA | 5,000 | - | NVIDIA | - |
| Meta GenAI cluster (RoCE) 2024-03 | Accelerator | USA | 24,576 | - | NVIDIA | - |
| Meta GenAI cluster (InfiniBand) 2024-03 | Accelerator | USA | 24,576 | - | NVIDIA | - |
| Tokyo-1 2024-02 | Accelerator | Japan | 128 | - | NVIDIA | - |
| Stampede3 2024-01 | Accelerator | USA | 96 | - | NVIDIA | - |
| Njoerd 2024 | Accelerator | United Kingdom | - | 78.20 PFlop/s | NVIDIA | #35 → |
| CHIE-3 2024 | Accelerator | Japan | - | 91.94 PFlop/s | NVIDIA | #28 → |
| ARF-ACC 2024 | Accelerator | Turkey | - | 12.09 PFlop/s | NVIDIA | #145 → |
| Opera Iceland KEF-1 SuperPOD 2024 | Accelerator | Iceland | - | 10.53 PFlop/s | NVIDIA | #88 → |
| Gautschi AI 2024 | Accelerator | USA | 160 | 7.26 PFlop/s | NVIDIA | #157 → |
| Unaizah-4 2024 | Accelerator | Saudi Arabia | - | 4.42 PFlop/s | NVIDIA | #225 → |
| Minerva 2024 | Accelerator | USA | - | 7.91 PFlop/s | NVIDIA | #147 → |
| Ubilink H100 Cluster 2024 | Accelerator | Taiwan | 1,024 | 45.82 PFlop/s | NVIDIA | #44 → |
| fortythree 2024 | Accelerator | Germany | 280 | 11.27 PFlop/s | NVIDIA | #42 → |
| SEAS H100 (Harvard) 2024 | Accelerator | USA | - | 3.38 PFlop/s | NVIDIA | #298 → |
| Medusa (Dell) 2024 | Accelerator | USA | - | 3.08 PFlop/s | NVIDIA | #387 → |
| MareNostrum 5 ACC 2023-12 | Accelerator | Spain | - | 138.2 PFlop/s | NVIDIA | #8 → |
| Eagle 2023-11 | Accelerator | USA | - | 561.2 PFlop/s | NVIDIA | #3 → |
| Eos 2023-11 | Accelerator | USA | - | 121.4 PFlop/s | NVIDIA | #23 → |
| Israel-1 2023-11 | Accelerator | Israel | 2,048 | 41.50 PFlop/s | NVIDIA | #34 → |
| Nabuchodonosor 2023-10 | Accelerator | France | 1,016 | - | NVIDIA | - |
| NHN Cloud Gwangju AI 2023-10 | Accelerator | South Korea | - | 12.84 PFlop/s | NVIDIA | #97 → |
| CoreWeave AI cluster for Inflection AI 2023-06 | Accelerator | USA | 3,584 | - | NVIDIA | - |
| wICE 2023-02 | Accelerator | Belgium | 20 | - | NVIDIA | - |
| Kempner AI Cluster 2023 | Accelerator | USA | 384 | 16.29 PFlop/s | NVIDIA | #85 → |
| DeepL Mercury 2023 | Accelerator | Sweden | - | 21.85 PFlop/s | NVIDIA | #34 → |
| ISEG 2023 | Accelerator | Netherlands | - | 46.54 PFlop/s | NVIDIA | #15 → |
| Yep1 2023 | Accelerator | USA | - | 23.32 PFlop/s | NVIDIA | #34 → |
| Pre-Eos 2023 | Accelerator | USA | 1,024 | 40.66 PFlop/s | NVIDIA | #14 → |
| Bridges-2 2021-03 | Accelerator | USA | 80 | - | NVIDIA | - |
| Phoenix (PACE) 2020 | Accelerator | USA | 32 | 1.84 PFlop/s | NVIDIA | #277 → |
| Sherlock 2014-01 | Accelerator | USA | - | - | NVIDIA | - |
| Snellius (Phase 3 GPU) | Accelerator | Netherlands | 352 | 13.64 PFlop/s | NVIDIA | #94 → |
| SAKURAONE | Accelerator | Japan | 800 | 33.95 PFlop/s | NVIDIA | #49 → |
| Santos Dumont | Accelerator | Brazil | 248 | 14.29 PFlop/s | NVIDIA | #52 → |
| Databricks DBRX Training Cluster | Accelerator | USA | 3,072 | - | NVIDIA | - |
| Della | Accelerator | USA | 336 | - | NVIDIA | - |
| farm22 | Accelerator | United Kingdom | 64 | - | NVIDIA | - |
| St. Jude HPC Research Cluster | Accelerator | USA | 112 | - | NVIDIA | - |
| BNY DGX SuperPOD | Accelerator | USA | - | - | NVIDIA | - |
| Freyja | Accelerator | Iceland | 248 | - | NVIDIA | - |
| CZI GPU Cluster | Accelerator | USA | 1,024 | - | NVIDIA | - |
| Kyber | Accelerator | France | 224 | - | NVIDIA | - |
| Tuwaiq-1 | Accelerator | Saudi Arabia | - | 8.49 PFlop/s | NVIDIA | #111 → |
| Nebius Mäntsälä Cluster | Accelerator | Finland | - | - | NVIDIA | - |
| Killarney | Accelerator | Canada | 80 | - | NVIDIA | - |
| Jetstream2 | Accelerator | USA | 96 | - | NVIDIA | - |
| Marlowe (Stanford) | Accelerator | USA | 248 | - | NVIDIA | - |
| Nano 5 | Accelerator | Taiwan | - | 13.06 PFlop/s | NVIDIA | #135 → |
| Taipei-1 | Accelerator | Taiwan | 512 | 22.30 PFlop/s | NVIDIA | #38 → |
| Hops | Accelerator | USA | - | 10.03 PFlop/s | NVIDIA | #142 → |
| Quest (Northwestern) | Accelerator | USA | 96 | - | NVIDIA | - |
| CUBOX Cray XD6500 | Accelerator | South Korea | - | 4.96 PFlop/s | NVIDIA | #155 → |
| DeepL Ceres | Accelerator | Sweden | - | 11.30 PFlop/s | NVIDIA | #107 → |
| Raphael | Accelerator | USA | - | 4.09 PFlop/s | NVIDIA | #244 → |
| Voltage Park H100 Cloud | Accelerator | USA | 24,000 | - | NVIDIA | - |
| Makau | Accelerator | USA | 64 | - | NVIDIA | - |
“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.