HPE Cray XD670 8× NVIDIA H200 GPU Servers: Special Offer
- Mar 30
- 3 min read
Updated: Apr 19
The HPE Cray XD670 offered here is a 5U enterprise AI server configured with 8× NVIDIA H200 141GB SXM GPUs, dual Intel Xeon Platinum 8562Y+, and 2TB DDR5 memory. This is a fixed, validated configuration available as a refurbished, fully tested, deployment-ready system.
HPE Cray XD670 with 8× NVIDIA H200
Limited stock at special pricing
Configuration Overview: HPE Cray XD670 8× NVIDIA H200 SXM Servers
This HPE Cray XD670 configuration is built around an 8-GPU NVIDIA HGX H200 141GB SXM platform fully interconnected with NVIDIA NVLink, designed for sustained AI training, distributed workloads, and GPU-accelerated HPC.
It is designed for:
Sustained multi-GPU AI training
Distributed model parallelism and data parallelism
Multi-node cluster scaling using high-speed fabric
The servers offered are based on a fixed, fully supported configuration selected for stability, compatibility, and real-world deployment.
Full Configuration: HPE Cray XD670 8× NVIDIA H200 SXM Servers
Base System: HPE Cray XD670 5U Rack Server
Component | Details | Why it matters |
Server / Chassis | HPE Cray XD670 (5U) | Purpose-built AI and HPC platform |
Form Factor | 5U high-density system | Optimized GPU density |
Architecture | Single node, dual CPU | Efficient compute + GPU balance |
PCIe | PCIe Gen5 support | High-bandwidth expansion |
GPUs: 8× NVIDIA HGX H200 141GB SXM
Component | Details | Why it matters |
GPUs | 8× NVIDIA HGX H200 141GB SXM | High-performance AI training |
GPU Memory | 141GB HBM3e per GPU | Supports very large models |
Interconnect | NVIDIA NVLink (fully interconnected) | Fast GPU-to-GPU communication |
Each GPU supports up to ~700W TDP.
The HGX SXM architecture enables:
Direct NVLink connectivity between GPUs
Reduced scaling overhead in multi-GPU training
Efficient collective communication
This platform is optimized for large transformer and foundation model workloads.
CPUs: 2× Intel Xeon Platinum 8562Y+
Component | Details | Why it matters |
CPUs | 2× Intel Xeon Platinum 8562Y+ | Host compute and orchestration |
Generation | 5th Gen Xeon Scalable | Latest platform support |
Role | Data + orchestration | Supports GPU pipeline efficiency |
The system supports dual 5th Gen Xeon CPUs up to high TDP levels, designed to feed 8 GPUs efficiently.
CPUs handle:
Data ingestion and preprocessing
Job scheduling and orchestration
Storage and network coordination
Memory: 2048GB DDR5 (32× 64GB DDR5-5600)
Component | Details | Why it matters |
Installed Memory | 2048GB (32× 64GB) | Large dataset handling |
DIMM Rating | DDR5-5600 | High bandwidth |
Slots | Up to 32 DIMMs | Maximum platform capacity |
The platform supports up to 32 DDR5 DIMMs at up to 5600 MT/s, depending on configuration.
Storage: NVMe OS Storage (M.2 RAID)
Component | Details | Why it matters |
OS Storage | 960GB NVMe SSD | Reliable OS / boot |
Adapter | Gigabyte PCIe M.2 RAID adapter | Redundant OS storage |
Type | NVMe Read Intensive | Fast boot and caching |
Local NVMe storage is typically used for:
OS and system partitions
Cache and staging
Metadata workloads
Networking: HPE InfiniBand NDR / Ethernet
Component | Details | Why it matters |
Adapter | HPE IB NDR200 / 200G QSFP112 | High-speed cluster interconnect |
Protocols | InfiniBand / Ethernet | Flexible deployment |
Speed | Up to 200Gb/s | Low-latency communication |
High-speed networking supports:
Distributed AI training
RDMA
Multi-node scaling
The HPE XD670 supports multiple interconnect options including InfiniBand and Ethernet fabrics.
Management
Component | Details | Why it matters |
Management | HPE Cray software + Redfish | Cluster lifecycle control |
Supports:
Remote provisioning
Monitoring
Firmware updates
Power & Cooling
Component | Details | Why it matters |
Power | High-density PSU configuration | Supports GPU-heavy load |
Cooling | Air or Direct Liquid Cooling (DLC) | Efficient thermal management |
GPU TDP | Up to ~700W each | Requires datacenter infrastructure |
The system supports both air cooling and direct liquid cooling, depending on deployment needs.
Platform Overview: HPE Cray XD670 with 8× NVIDIA H200
The HPE Cray XD670 is engineered for AI infrastructure environments where:
GPU density
Thermal efficiency
High-speed networking
Scalable cluster design
are critical for production workloads.
It is designed to scale from single-node deployments to large multi-node AI clusters.
Use Cases: HPE Cray XD670 with 8× NVIDIA H200 GPUs
Large language model training
Foundation model development
Multi-node distributed AI training
HPC simulation workloads
Scientific computing
Testing, Condition, and Warranty: HPE Cray XD670 with 8× NVIDIA H200 GPUs
Condition:
Refurbished, fully tested
Testing includes:
CPU, memory, and GPU validation
NVMe and storage verification
Network adapter testing
Stability testing under load
Warranty:
3-Year hardware warranty included
Fully tested before delivery
Deployment-ready configuration
Extended support options available upon request.
HPE Cray XD670 with 8× NVIDIA H200
Limited stock at special pricing
Comments