Parts/Accelerators/NVIDIA B200 SXM
Accelerators · Blackwell · 2024
NVIDIA B200 SXM
192 GB HBM3E SXM module. The unit most 2025-era commercial clusters are counted in, and almost none of them submit HPL results.
Why this part
Two reticle-limited dies joined by a 10 TB/s link and presented to software as one device. A single die can no longer be made larger, so the only remaining direction is to package more of them together and hide the seam, which is the same conclusion AMD reached with chiplets and Cerebras reached by abandoning the reticle limit entirely. At 1000W it is a liquid-cooling requirement rather than a preference.
Systems using this part
- Systems
- 31
- Rmax underneath
- 478.7 PFlop/s
- Units recorded
- 349,664
| System | Role | Country | Units | System Rmax | Supplied by | Rank |
|---|---|---|---|---|---|---|
| Firebird Armenian AI Factory (DC-1) 2026-08-08 | Accelerator | Armenia | 6,144 | - | NVIDIA | - |
| NIPA(NHN Cloud)-CL-1 2026-04 | Accelerator | South Korea | - | 137.4 PFlop/s | NVIDIA | #20 → |
| NIPA(NHN Cloud)-CL-2 2026-04 | Accelerator | South Korea | - | 68.42 PFlop/s | NVIDIA | #40 → |
| Industrial AI Cloud 2026-02-04 | Accelerator | Germany | 10,000 | - | NVIDIA | - |
| Viettel DGX B200 System 2026-02 | Accelerator | Vietnam | - | - | NVIDIA | - |
| Jairu 2026-02 | Accelerator | Brazil | 96 | - | NVIDIA | - |
| NIPA (Kakao) 2026-01 | Accelerator | South Korea | - | 69.10 PFlop/s | NVIDIA | #39 → |
| Naver B200 4K Cluster 2026-01 | Accelerator | South Korea | 4,000 | - | NVIDIA | - |
| SEIRAN 2025-10 | Accelerator | Japan | 96 | 3.32 PFlop/s | NVIDIA | #374 → |
| Nebius Modiin Cluster 2025-10 | Accelerator | Israel | 4,000 | - | NVIDIA | - |
| OpenCanvas GPUaaS SP1 2025-10 | Accelerator | Japan | - | 17.53 PFlop/s | NVIDIA | #105 → |
| SDS-AI 2025-09 | Accelerator | Israel | 2,032 | - | NVIDIA | - |
| The F.I.N 2025-09 | Accelerator | Poland | - | 7.57 PFlop/s | NVIDIA | #211 → |
| Haein Cluster 2025-08-05 | Accelerator | South Korea | 1,000 | - | NVIDIA | - |
| CHIE-4 2025-07 | Accelerator | Japan | - | 135.4 PFlop/s | NVIDIA | #21 → |
| Mayo Clinic DGX SuperPOD 2025-07 | Accelerator | USA | 128 | - | NVIDIA | - |
| Betty 2025-05 | Accelerator | USA | 248 | 8.49 PFlop/s | NVIDIA | #189 → |
| Minerva 2024 | Accelerator | USA | 48 | 7.91 PFlop/s | NVIDIA | #147 → |
| Microsoft Goodyear AI Datacenter 2023-04 | Accelerator | USA | 58,100 | - | NVIDIA | - |
| Microsoft Project Osmium 2021-01 | Accelerator | USA | 45,500 | - | NVIDIA | - |
| HiPerGator AI 2021 | Accelerator | USA | 504 | 17.49 PFlop/s | NVIDIA | #106 → |
| Solstice | Accelerator | USA | 100,000 | - | NVIDIA | - |
| Equinox | Accelerator | USA | 10,000 | - | NVIDIA | - |
| Project Ceiba | Accelerator | USA | 20,736 | - | NVIDIA | - |
| E2E Chennai B200 Cluster | Accelerator | India | 1,024 | - | NVIDIA | - |
| Meta Rosemount (Minnesota) | Accelerator | USA | 85,400 | - | NVIDIA | - |
| AICR | Accelerator | USA | 248 | - | NVIDIA | - |
| Bouchet | Accelerator | USA | 56 | - | NVIDIA | - |
| Hopper (Yale) | Accelerator | USA | 48 | - | NVIDIA | - |
| DAIS | Accelerator | Germany | 128 | 6.03 PFlop/s | NVIDIA | #247 → |
| Skipjack (JHU) | Accelerator | USA | 128 | - | 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.