NVIDIA DGX SuperPOD with GB200 NVL72
Brand: NVIDIA | Category: Servers
SKU: DGX-SPOD-GB200 | Part #: DGX-SPOD-GB200 | MPN: DGX-SPOD-GB200
Contact for Pricing — Request a Quote
About the NVIDIA DGX SuperPOD with GB200 NVL72
The NVIDIA DGX SuperPOD with GB200 NVL72 (part number DGX-SPOD-GB200) connects directly via NVIDIA Quantum-2 InfiniBand at 400 Gb/s per GPU, delivering enterprise-grade network fabric for large-scale AI workloads with seamless all-to-all communication across the rack-scale system.
This rack-scale system houses 72x NVIDIA B200 Blackwell GPUs paired with 36x NVIDIA Grace CPU modules—each delivering 72 Arm Neoverse V2 cores—interconnected by 5th Generation NVLink at 900 GB/s bidirectional throughput per Grace-Blackwell pair. The NVL72 chassis aggregates 1.8 TB/s of NVLink fabric bandwidth and delivers up to 1.44 exaflops of peak AI compute (FP4), with each B200 GPU contributing up to 20 petaflops. Memory capacity reaches 13.8 TB of HBM3e across the rack, with each GPU supplying 192 GB at 8 TB/s bandwidth. The platform supports FP64, TF32, BF16, FP16, FP8, and FP4 numerical precisions, running NVIDIA's full software stack including CUDA, cuDNN, TensorRT, NVIDIA NeMo, and NVIDIA AI Enterprise. Direct liquid cooling comes standard.
AI infrastructure teams and enterprise ML operations centers deploy the DGX-SPOD-GB200 to train large language models, conduct multimodal research, and run inference clusters at datacenter scale. The system supports PyTorch, JAX, and TensorFlow frameworks, enabling seamless integration into existing ML pipelines. Organizations requiring massive parallel GPU compute with unified memory architecture and low-latency interconnect find this platform essential for production-grade AI workloads. Contact Omnixon Global to request a quotation for the NVIDIA DGX SuperPOD with GB200 NVL72.
```Technical Specifications
| Brand | NVIDIA |
| Category | Servers |
| SKU | DGX-SPOD-GB200 |
| Part Number | DGX-SPOD-GB200 |
| Condition | New |
| Manufacturer Part Number | DGX-SPOD-GB200 |
| Product Line | DGX SuperPOD |
| Form Factor | Rack-scale system (NVL72 chassis) |
| GPUs per NVL72 Rack | 72x NVIDIA B200 Blackwell GPUs |
| GPU Architecture | NVIDIA Blackwell |
| CPU Modules per NVL72 Rack | 36x NVIDIA Grace CPU modules (72 Arm Neoverse V2 cores each) |
| CPU-GPU Interconnect | NVLink-C2C at 900 GB/s per Grace-Blackwell pair (bidirectional) |
| NVLink Generation | 5th Generation NVLink |
| NVLink Fabric Bandwidth per Rack | 1.8 TB/s aggregate (all-to-all within NVL72) |
| Peak AI Compute per B200 GPU | Up to 20 petaflops (FP4 Tensor Core) |
| Peak AI Compute per NVL72 Rack | Up to 1.44 exaflops (FP4) |
| GPU Memory per B200 | 192 GB HBM3e |
| Total GPU Memory per NVL72 Rack | 13.8 TB HBM3e |
| GPU Memory Bandwidth per B200 | 8 TB/s |
| Supported Numerical Precisions | FP64, TF32, BF16, FP16, FP8, FP4 |
| Network Fabric | NVIDIA Quantum-2 InfiniBand, 400 Gb/s per GPU |
| Cooling | Direct liquid cooling (standard) |
| Software Stack | CUDA, cuDNN, TensorRT, NVIDIA NeMo, NVIDIA AI Enterprise |
| Supported AI Frameworks | PyTorch, JAX, TensorFlow |
| Scalability | Multi-rack SuperPOD scaling via InfiniBand fabric |
| Target Deployment | Enterprise datacenter, HPC, sovereign AI, cloud on-premises |
Frequently Asked Questions about NVIDIA DGX SuperPOD with GB200 NVL72
What does the NVIDIA DGX SuperPOD with GB200 NVL72 do?
The NVIDIA DGX SuperPOD with GB200 NVL72 is built for enterprise data-center workloads — virtualization (VMware, Proxmox, Nutanix), private cloud, database hosting, and AI/ML training. It fits standard EIA-310 server racks and supports redundant PSUs and hot-swap drives common in production environments.
What are the headline specs of the NVIDIA DGX SuperPOD with GB200 NVL72?
Key specifications for the NVIDIA DGX SuperPOD with GB200 NVL72: new condition; gpu support NVIDIA GB200; ai optimized Yes. Manufacturer part number DGX-SPOD-GB200. For the full datasheet with electrical, environmental, and compliance details, contact our pre-sales engineering team.