NVIDIA GB200 NVL72 liquid-cooled rack-scale platform with 72 Blackwell GPUs, 36 Grace CPUs, 13.4 TB HBM3e, and a 130 TB/s NVLink domain.
Fast Shipping
100k+ SKUs
8/5 Support
Certified
NVIDIA GB200-NVL72 | Grace Blackwell Rack-Scale AI Platform
NVIDIA GB200 NVL72 Grace Blackwell Rack-Scale AI Platform (GB200-NVL72)
NVIDIA GB200 NVL72 is a liquid-cooled rack-scale Grace Blackwell platform that connects 72 Blackwell GPUs and 36 Grace CPUs into a large NVLink compute domain. The architecture is designed for real-time large-model inference, large-scale training, accelerated data processing, and high-performance computing.
Availability / RFQ: This product is a current NVIDIA platform or component and is listed for RFQ. Pricing, allocation, quantity, configuration, support, and lead time can vary with current channel conditions. Contact Catalyst Data Solutions to discuss current status, compatibility, configuration requirements, and sourcing options.End-user / export
Compliance notice: Advanced AI accelerators and systems may be subject to applicable U.S. export controls, end-user/end-use restrictions, destination requirements, and OEM/channel policies. Catalyst may request purchaser, final end-user, intended-use, and deployment-location information before a quote or fulfillment can be approved. An RFQ does not guarantee allocation or shipment.
Technical Specifications
| SKU | GB200-NVL72 |
|---|---|
| Product Type | Rack-Scale Grace Blackwell AI Platform |
| GPU / CPU Configuration | 72 Blackwell GPUs / 36 Grace CPUs |
| GPU Memory | 13.4 TB HBM3e |
| GPU Memory Bandwidth | 576 TB/s aggregate |
| CPU Memory | 17 TB LPDDR5X |
| NVLink Bandwidth | 130 TB/s |
| Cooling | Liquid-cooled rack-scale design |
| Scale-Out | Quantum-X800 InfiniBand or Spectrum-X800 Ethernet |
| Availability Status | Available / RFQ |
Deployment and Compatibility Considerations
- Validate rack, CDU, cooling water, power distribution, and facility integration.
- Select scale-out fabric, ConnectX/BlueField connectivity, optics, cables, and switch topology for the target pod size.
- Plan storage throughput, checkpoint architecture, dataset ingestion, and management networks.
- Confirm the chosen OEM implementation and exact service/support model.
Common Use Cases
- Large-model inference
- Generative AI training
- Hyperscale AI factories
- Accelerated data processing
- HPC and AI convergence
Frequently Paired Infrastructure
- Quantum-X800 InfiniBand
- Spectrum-X800 Ethernet
- BlueField-3 DPUs
- AI storage
- Liquid-cooled rack infrastructure
Official NVIDIA Documentation
Frequently Asked Questions
How many GPUs and CPUs are in GB200 NVL72?
The platform combines 72 Blackwell GPUs and 36 Grace CPUs.
How much GPU memory does GB200 NVL72 provide?
NVIDIA lists 13.4 TB of aggregate HBM3e GPU memory.
| product type | Rack-Scale AI Platform |
|---|---|
| Architecture / Series | NVIDIA Grace Blackwell GB200 |
| Accelerator Configuration | 72 Blackwell GPUs / 36 Grace CPUs |
| Cooling Type | Liquid-Cooled Rack-Scale |
| Availability Status | Available / RFQ |
