Nvidia GB200 NVL72
The Nvidia GB200 NVL72 is a liquid-cooled rack joining 36 Grace CPUs and 72 Blackwell GPUs in one NVLink domain.
- Vendor
- Nvidia
- Architecture
- Grace Blackwell
- Memory
- 13.4 TB HBM3E (72 GPUs)
- Memory bandwidth (TB/s)
- 576 TB/s
- FP8 (TFLOPS)
- 720,000 TFLOPS
- FP4 (PFLOPS)
- 720 PFLOPS
- Form factor
- Rack-scale system: 72 GPUs, 36 Grace CPUs
- Status
- Shipping
GB200 NVL72 has 130 TB/s of GPU-to-GPU bandwidth. It is built from 36 GB200 Grace Blackwell Superchips, each pairing one Grace CPU with two Blackwell GPUs (372 GB HBM3E, 16 TB/s, 40 sparse / 20 dense PFLOPS NVFP4). The figures in this row are for the whole rack: 13.4 TB of HBM3E, 576 TB/s and 720 PFLOPS of dense NVFP4.
- Memory (GB)
- 13,400 GB
- Precision note (sparse or dense)
- Rack totals. FP8/FP6 720 PFLOPS is sparse (Nvidia: dense is one-half). NVFP4 1,440 PFLOPS sparse / 720 PFLOPS dense.
- Vendor source
- nvidia.com
- Checked
Questions
How many GPUs are in a GB200 NVL72?
A GB200 NVL72 rack has 72 Blackwell GPUs and 36 Grace CPUs in one NVLink domain.
What is a GB200 Grace Blackwell Superchip?
Each GB200 Superchip pairs one Grace CPU with two Blackwell GPUs, with 372 GB of HBM3E and 16 TB/s.
How much memory does a GB200 NVL72 rack have?
The rack has 13.4 TB of HBM3E across 72 GPUs, with 576 TB/s of memory bandwidth.
What is the GB200 NVL72’s FP4 performance?
Nvidia lists NVFP4 at 1,440 PFLOPS sparse and 720 PFLOPS dense per rack; FP8/FP6 is 720 PFLOPS sparse.
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.