HPE Cray XD680 (NVIDIA Blackwell GB200 NVL72 Configuration)
Brand: HPE | Category: GPUs
SKU: HPE-R9W12A | Part #: R9W12A | MPN: R9W12A
Contact for Pricing — Request a Quote
About the HPE Cray XD680 (NVIDIA Blackwell GB200 NVL72 Configuration)
HPE Cray XD680 with NVIDIA Blackwell GB200 NVL72 configuration delivers exceptional AI training and inference performance through a unified architecture combining 72 Blackwell B200 GPU dies with 36 NVIDIA Grace CPU modules. Built for enterprise AI infrastructure teams seeking rack-scale GPU density, this system integrates HPE's proven Cray XD Series platform with next-generation accelerator technology. The GB200 NVL72 module provides 720 GB of HBM3e memory and achieves up to 1.4 ExaFLOPS of FP4 compute performance at rack scale, enabling rapid model training and deployment across diverse AI workloads.
The XD680 (part number R9W12A) employs NVLink 5th Generation for 1.8 TB/s aggregate GPU-to-GPU bandwidth and NVLink-C2C interconnect delivering 900 GB/s per Grace-Blackwell pair, ensuring minimal data movement bottlenecks. Direct Liquid Cooling (DLC) maintains thermal efficiency in dense deployments, while HPE Slingshot 200Gb/s high-speed interconnect fabric enables seamless multi-rack scaling. Management through HPE iLO and native support for Linux (RHEL, Ubuntu), NVIDIA AI Enterprise stack, and CUDA 12.x and later simplify integration into existing enterprise environments. IT procurement teams and AI infrastructure architects can standardize on a single rack-optimized platform that balances compute density, memory bandwidth, and operational simplicity.
Typical AI and HPC Workloads
- Large language model (LLM) training and fine-tuning at scale with reduced training iteration cycles
- Generative AI inference serving with high throughput and low-latency response requirements
- Scientific computing and computational fluid dynamics simulations requiring sustained multi-GPU performance
- Financial modeling, risk analysis, and real-time analytics on massive datasets
- Multi-tenant AI platform consolidation within a single rack-scale deployment
To discuss HPE Cray XD680 specifications, integration requirements, and procurement, contact Omnixon Global for a detailed RFQ.
```Technical Specifications
| Brand | HPE |
| Category | GPUs |
| SKU | HPE-R9W12A |
| Part Number | R9W12A |
| Condition | New |
| Form Factor | Rack-Mount |
| Interface | PCIe Gen5 |
| Manufacturer Part Number | R9W12A |
| Product Line | HPE Cray XD Series |
| Model | HPE Cray XD680 |
| GPU Architecture | NVIDIA Blackwell |
| GPU Configuration | GB200 NVL72 (72x Blackwell B200 GPU dies + 36x Grace CPU modules) |
| Total GPUs per NVL72 Module | 72 Blackwell B200 GPU dies |
| Total CPUs per NVL72 Module | 36 NVIDIA Grace (Arm Neoverse V2) |
| GPU Interconnect | NVLink 5th Generation, 1.8 TB/s aggregate GPU-to-GPU bandwidth |
| CPU-GPU Interconnect | NVLink-C2C (chip-to-chip), 900 GB/s per Grace-Blackwell pair |
| Total HBM Memory (NVL72) | 720 GB HBM3e |
| AI Compute Performance (FP4) | Up to 1.4 ExaFLOPS (NVL72 rack scale) |
| CPU Memory Type | LPDDR5X (per Grace module) |
| Cooling | Direct Liquid Cooling (DLC) |
| Network Fabric Support | HPE Slingshot 200Gb/s high-speed interconnect |
| Management | HPE iLO |
| Form Factor | Rack-scale module (fits HPE Cray XD rack infrastructure) |
| Operating System Support | Linux (RHEL, Ubuntu); NVIDIA AI Enterprise stack |
| CUDA Support | CUDA 12.x and later |
Frequently Asked Questions about HPE Cray XD680 (NVIDIA Blackwell GB200 NVL72 Configuration)
What does the HPE Cray XD680 (NVIDIA Blackwell GB200 NVL72 Configuration) do?
The HPE Cray XD680 (NVIDIA Blackwell GB200 NVL72 Configuration) accelerates AI/ML training, inference, scientific HPC, and virtualization (vGPU) workloads. Typical deployments include LLM training clusters, computer-vision pipelines, financial risk modeling, and rendering farms.
What are the headline specs of the HPE Cray XD680 (NVIDIA Blackwell GB200 NVL72 Configuration)?
Key specifications for the HPE Cray XD680 (NVIDIA Blackwell GB200 NVL72 Configuration): form factor Rack-Mount; interface PCIe Gen5; 3 years support; new condition; support 3 years standard (manufacturer); gpu support NVIDIA GB200; ai optimized Yes. Manufacturer part number R9W12A. For the full datasheet with electrical, environmental, and compliance details, contact our pre-sales engineering team.
Is the HPE Cray XD680 (NVIDIA Blackwell GB200 NVL72 Configuration) compatible with my infrastructure?
The HPE Cray XD680 (NVIDIA Blackwell GB200 NVL72 Configuration) requires a PCIe Gen4 or Gen5 x16 slot, server power adequate for the card's TDP, and CUDA/ROCm driver support in your hypervisor or bare-metal OS. Sales engineering will confirm chassis fit (1U/2U/4U), PCIe lane count, and PSU headroom before quoting.