NVIDIA GB200 NVL72
GB200 NVL72 puts 72 Blackwell GPUs in one liquid-cooled rack. We build new clusters of them and take over racks that are already installed.
- GPUs
- 72 GPUs, 36 Grace CPUs
- GPU memory
- 13.4 TB HBM3e per rack
- NVLink
- 130 TB/s across 72 GPUs
- Power
- About 120 kW per rack
- Cooling
- Liquid, at the rack
NVIDIA GB200 NVL72 specifications.
The rack as NVIDIA publishes it, and what each figure means on site.
- Generation
- BlackwellGB300 NVL72 keeps the same rack layout on Blackwell Ultra.
- GPUs
- 72 GPUs per rackOne NVLink domain, so the rack arrives, installs and runs as one system.
- CPUs
- 36 Grace per rackEach Grace CPU is paired with two GPUs.
- GPU memory
- 186 GB HBM3e per GPU, 13.4 TB per rackA model split across all 72 GPUs has the full 13.4 TB to work with over NVLink.
- NVLink
- 130 TB/s across 72 GPUs1.8 TB/s per GPU. Traffic inside the rack stays off the scale-out fabric.
- Scale-out network
- 400 Gb/s per GPU (ConnectX-7)A rack has 72 of them, 28.8 Tb/s in total.
- Power
- About 120 kW per rackThe rack alone. The fabric and storage add their own load.
- Cooling
- Liquid, at the rackLiquid takes most of the heat. The rest leaves as air.
- Loaded weight
- About 1.4 to 1.6 metric tonsFrom server manufacturers' documents. It varies with the build.
- Reference design
- 8 racksNVIDIA's scalable unit, 576 GPUs.
- Lead time
- Set in the proposalRacks already installed skip this. A takeover starts with an assessment.
Figures are NVIDIA's for the GB200 NVL72 rack, except loaded weight, which comes from server manufacturers' documents.
What a rack position needs before power-on.
One rack in elevation, numbered to the notes. On racks already installed, these are the four points a takeover assessment starts from.
- 1
About 120 kW per rack
The rack's power shelves take two independent feeds. We size both for peak draw and prove them under load during burn-in.
- 2
A coolant loop commissioned first
A coolant distribution unit, in the rack or in the row, feeds cold plates on every compute tray. The loop is flushed, filled and held at test pressure, and every quick disconnect is checked, before a tray powers on.
- 3
72 ConnectX-7 links to the leaf switches
One per GPU at 400 Gb/s, on a spine-leaf RoCE v2 Ethernet fabric. They carry the traffic between racks; NVLink handles everything within one.
- 4
About 1.4 to 1.6 metric tons loaded
All of it on one rack footprint. We check the floor rating at each position, and the route in from the loading dock, before the racks ship.
GB200 NVL72 against GB300 NVL72.
Both put 72 GPUs in one NVLink domain per rack. GB300 NVL72 adds memory and network per GPU, at more power per rack.
| GB300 NVL72 | GB200 NVL72 | |
|---|---|---|
| Generation | Blackwell Ultra | Blackwell |
| GPU memory | 288 GB HBM3e per GPU | 186 GB HBM3e per GPU |
| Memory per rack | 20.7 TB per rack | 13.4 TB per rack |
| Scale-out network | 800 Gb/s per GPU | 400 Gb/s per GPU |
| Power | Up to 142 kW per rack | About 120 kW per rack |
| Reference designs | 2, 4 and 8 racks | 8 racks |
| Fits best | New builds, and models that need the most memory per GPU | Models that fit its memory, and installed racks that need an operator |
Figures from NVIDIA. Either is quoted per project, built new or taken over.
Other NVIDIA platforms we build on.
The same four stages on each: design, source, build and operate. All hardware.
- NVIDIA HGX B300Blackwell Ultra8 GPUs per node288 GB HBM3e per GPUAirAvailable
- NVIDIA HGX B200Blackwell8 GPUs per node180 GB HBM3e per GPUAirAvailable
- NVIDIA GB300 NVL72Blackwell Ultra72 GPUs per rack288 GB HBM3e per GPULiquidAvailable
- NVIDIA HGX Rubin NVL8Rubin8 GPUs per node288 GB HBM4 per GPULiquidWaitlist
- NVIDIA VR200 NVL72Rubin72 GPUs per rack288 GB HBM4 per GPULiquidWaitlist
GB200 NVL72 questions.
We build GB200 NVL72 clusters as a deployment your team runs, a managed cluster you own and we operate, or a reserved cluster we own. Other platforms.
NVIDIA gives about 120 kW per rack. That covers the rack only; leaf switches and storage are fed separately.
Facility water at the position and a coolant distribution unit, in the rack or in the row. Part of the heat still leaves as air, so the hall keeps its air handling.
We flush and fill the loop, hold it at test pressure and check every quick disconnect for leaks. Trays power on only after that, and the flow, temperature and leak sensors go into our monitoring.
Yes, as a managed cluster: you keep ownership and we operate it on our software. We assess the racks, the loop and the firmware first, fix what needs fixing, then run it to an SLA.
Server manufacturers' documents put a loaded rack at about 1.4 to 1.6 metric tons. On a takeover, we confirm the floor under each installed rack is rated for it.
One rack, 72 GPUs. NVIDIA's scalable unit is 8 racks, or 576 GPUs.
It can be. GB300 NVL72 raises memory to 288 GB per GPU from 186, and the network to 800 Gb/s per GPU from 400 Gb/s, at more power per rack. GB200 NVL72 suits workloads that fit in 186 GB per GPU, on positions fed for about 120 kW per rack.