(858) 613-1818Quote list 0
Coming soon

NVIDIA Vera Rubin NVL72 rack

NVIDIA's Rubin generation rack with 72 Rubin GPUs, 36 Vera CPUs, 20.7 TB of HBM4 and 3,600 PFLOPS of sparse NVFP4 inference. Register your interest now.

GPU: NVIDIA Vera Rubin NVL72

Illustration of a rack of servers
GPUs
72 x NVIDIA Rubin, 288 GB HBM4 each, 20.7 TB per rack at 1,400 TB/s
CPUs
36 x NVIDIA Vera, 3,168 custom Olympus cores, up to 54 TB LPDDR5X
NVLink
Sixth generation, 216 TB/s per rack, 65 TB/s NVLink-C2C
Performance
3,600 PFLOPS sparse NVFP4 inference, 2,520 PFLOPS dense NVFP4 training, per NVIDIA
Scale out
ConnectX-9 SuperNICs and BlueField-4 DPUs, 32.4 TB/s bi-directional per rack
Power and cooling
Liquid cooled; per GPU and rack power not yet published

Overview

Vera Rubin NVL72 is the successor to GB300 NVL72 and the first rack scale system of NVIDIA's Rubin generation. Published figures give 72 Rubin GPUs with 20.7 TB of HBM4 at 1,400 TB/s, 36 Vera CPUs with 3,168 custom Olympus cores, sixth generation NVLink at 216 TB/s, and 3,600 PFLOPS of sparse NVFP4 inference per rack, with ConnectX-9 and BlueField-4 for scale out.

NVIDIA states that Vera Rubin is ramping into full production with its system partners, but per GPU power, rack power and reseller delivery dates are not yet published, so we list the rack as coming soon and take expressions of interest rather than firm orders. A Rubin rack will need at least the liquid cooling and power density of GB300 NVL72, which is a useful planning baseline now.

If you are planning a Rubin deployment, we can reserve a queue position, run the site survey against the GB300 NVL72 requirements, and quote an interim GB300 NVL72 or HGX B300 cluster so that the facility and the software stack are ready when Rubin ships.

Use cases

  • Planning next generation AI factories
  • Roadmap alignment for sovereign compute
  • Reasoning model inference at extreme scale

Specifications

Compute
Configuration72 x Rubin GPUs and 36 x Vera CPUs, per NVIDIA
GPU memory20.7 TB HBM4, 1,400 TB/s
CPU memoryUp to 54 TB LPDDR5X
NVFP4 performance3,600 PFLOPS sparse inference, 2,520 PFLOPS dense training
FP8 performance1,260 PFLOPS dense
Interconnect
NVLinkSixth generation NVLink, 216 TB/s per rack
NVLink-C2C65 TB/s per rack
Scale outConnectX-9 SuperNICs and BlueField-4 DPUs, Quantum-X800 InfiniBand or Spectrum-X Ethernet
Facility
Form factorSingle rack, 72 GPU NVLink domain
CoolingLiquid cooled, 45 C inlet temperature per NVIDIA; CDU and facility water loop required
PowerNot yet published
AvailabilityNVIDIA states Vera Rubin is ramping into full production; reseller delivery dates not yet published

Figures describe the system class. Exact configuration, weights and power draw are confirmed on the quotation.

What's included

  • Expression of interest and queue position
  • Site survey against GB300 NVL72 liquid cooling and power requirements
  • Specification updates as NVIDIA publishes them
  • Firm quotation when configurations are released

Compliance

Compare it with the next class up or down

The nearest classes we quote are the NVIDIA GB300 NVL72 rack and the 8 GPU HGX Rubin NVL8 AI server. If neither is quite right, tell us the workload and we specify the system between them.

Talk to an engineerRequest a quotation