Nvidia GB300 NVL72
The Nvidia GB300 NVL72 is the Blackwell Ultra version of Nvidia’s liquid-cooled rack-scale system.
- Vendor
- Nvidia
- Architecture
- Grace Blackwell Ultra
- Memory
- 20 TB HBM3E (72 GPUs)
- Memory bandwidth (TB/s)
- 576 TB/s
- FP8 (TFLOPS)
- 720,000 TFLOPS
- FP4 (PFLOPS)
- 1,080 PFLOPS
- Form factor
- Rack-scale system: 72 GPUs, 36 Grace CPUs
- Status
- Shipping
GB300 NVL72 combines 72 Blackwell Ultra GPUs and 36 Grace CPUs, with 20 TB of GPU memory and 37 TB of fast memory in total. Nvidia’s page marks it available now, and Nvidia names Microsoft, CoreWeave and Oracle Cloud Infrastructure among the clouds deploying it. The figures in this row are for the whole rack, with bandwidth stated as up to 576 TB/s.
- Memory (GB)
- 20,000 GB
- Precision note (sparse or dense)
- Rack totals. Nvidia: all Tensor Core specs are with sparsity unless noted. FP8/FP6 720 PFLOPS sparse; FP4 1,440 PFLOPS sparse / 1,080 PFLOPS without sparsity. Bandwidth is stated as up to 576 TB/s.
- Vendor source
- nvidia.com
- Checked
Questions
How much memory does the GB300 NVL72 have?
The rack has 20 TB of GPU memory and 37 TB of fast memory in total.
What is the GB300 NVL72’s FP4 performance?
Nvidia lists FP4 at 1,440 PFLOPS sparse and 1,080 PFLOPS without sparsity per rack; FP8/FP6 is 720 PFLOPS sparse.
Who is deploying the GB300 NVL72?
Nvidia names Microsoft, CoreWeave and Oracle Cloud Infrastructure among the clouds deploying it.
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.