Nvidia Vera Rubin NVL72
The Nvidia Vera Rubin NVL72 is Nvidia’s third-generation MGX rack, with 72 Rubin GPUs and 36 Vera CPUs.
- Vendor
- Nvidia
- Architecture
- Vera Rubin
- Memory
- 20.7 TB HBM4 (72 GPUs)
- Memory bandwidth (TB/s)
- 1,580 TB/s
- FP8 (TFLOPS)
- 1,260,000 TFLOPS
- FP4 (PFLOPS)
- 2,520 PFLOPS
- Form factor
- Rack-scale system: 72 GPUs, 36 Vera CPUs
- Status
- In production
The GPUs and CPUs in Vera Rubin NVL72 are linked by NVLink 6, with 20.7 TB of HBM4 and 54 TB of LPDDR5X. Nvidia announced full production on May 31, 2026, saying production shipments would begin in the fall, and by July 21, 2026 reported racks running at CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius. On September 30, 2026, CoreWeave said it had made Vera Rubin NVL72 available at production scale, with Cognition as the first customer; see our news report. The figures in this row are rack totals and preliminary per Nvidia.
- Memory (GB)
- 20,700 GB
- Precision note (sparse or dense)
- Rack totals. FP8/FP6 1,260 PFLOPS and NVFP4 2,520 PFLOPS are training figures marked dense; NVFP4 inference is listed as 3,600 PFLOPS without a sparse/dense label.
- Launch
- Launched January 5, 2026 (CES); full production announced May 31, 2026
- Vendor source
- nvidia.com
- Checked
Questions
How much memory does the Vera Rubin NVL72 have?
What is the Vera Rubin NVL72’s performance?
Who is running Vera Rubin NVL72?
When did Vera Rubin NVL72 enter production?
Source
Every figure is taken from the vendor’s own datasheet, product page, technical documentation or announcement, linked on each chip with the date it was last checked. How we pick and check sources: editorial standards.