Parts/Interconnect/NVIDIA InfiniBand XDR (Quantum-X800)
Interconnect · XDR · 2024
NVIDIA InfiniBand XDR (Quantum-X800)
800Gb/s per port with in-network compute. The scale-out half of the fabric where NVLink is the scale-up half.
Why this part
Quantum-X800 doubles the port rate again and keeps in-network reduction. Its role is scale-out, between racks, where NVLink is scale-up inside one. Reading the two together is the whole architecture of a modern AI system: a small number of accelerators joined tightly enough to look like one device, and then as many of those as the building can power, joined loosely.
Systems using this part
- Systems
- 7
- Rmax underneath
- 19.80 PFlop/s
- Units recorded
- -
0 of 7 edges disclose a quantity
| System | Role | Country | Units | System Rmax | Supplied by | Rank |
|---|---|---|---|---|---|---|
| ROQUO 2026-06-19 | Interconnect | Japan | - | 19.80 PFlop/s | NVIDIA | - |
| Azure NDv6 GB300 Cluster 2025-10 | Interconnect | USA | - | - | NVIDIA | - |
| Doudna | Interconnect | USA | - | - | NVIDIA | - |
| HammerHAI | Interconnect | Germany | - | - | NVIDIA | - |
| IT4LIA | Interconnect | Italy | - | - | NVIDIA | - |
| Yotta D2 Blackwell Ultra Cluster | Interconnect | India | - | - | NVIDIA | - |
| RIKYU | Interconnect | Japan | - | - | 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.