Parts/Interconnect/NVIDIA NVLink 5 / NVSwitch
Interconnect · NVLink 5 · 2024
NVIDIA NVLink 5 / NVSwitch
1.8TB/s per GPU across up to 72 GPUs in one NVLink domain. The intra-rack fabric that makes GB200 NVL72 a single scale-up unit rather than 72 nodes.
Why this part
1.8 TB/s per GPU is roughly fourteen times what a PCIe 5.0 x16 slot offers, and it is the number that makes GB200 NVL72 a single scale-up unit rather than 72 machines. NVLink is absent from most fabric-share analyses because it sits inside the node and no spec string names it, which understates NVIDIA's networking position considerably: the company supplies both the fabric between racks and the far faster one inside them.
Systems using this part
- Systems
- 1
- Rmax underneath
- 0.0 GFlop/s
- Units recorded
- -
| System | Role | Country | Units | System Rmax | Supplied by | Rank |
|---|---|---|---|---|---|---|
| Hon Hai Kaohsiung Super Computing Center | Interconnect | Taiwan | - | - | 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.