8-GPU SXM platform
Purpose-built for large-model AI training, deep learning and advanced HPC simulation.
HPE HPE Cray XD670
$150,000
HPE HPE Cray XD670 is a HPE 5U GPU-accelerated server designed for AI training, large-model inference, accelerated analytics and high-performance computing. It provides a configurable enterprise server CPU platform with enterprise ECC memory, high-speed storage, networking and management options, plus accelerator-oriented expansion for GPU-intensive workloads. Final CPU, memory, storage, accelerator and network selections should be matched to the deployment target.
Explore Related
Related products, categories and research.
AI & data-center infrastructure
HPE HPE Cray XD670
HPE HPE Cray XD670 is a HPE 5U GPU-accelerated server designed for AI training, large-model inference, accelerated analytics and high-performance computing. It provides a configurable enterprise server CPU platform with enterprise ECC memory, high-speed storage, networking and management options, plus accelerator-oriented expansion for GPU-intensive workloads. Final CPU, memory, storage, accelerator and network selections should be matched to the deployment target.
8×SXM GPUs
5UChassis
DLCCooling option
Product overview
Built around the real deployment workflow
Enterprise servers are configuration-driven products. The model defines the chassis and platform capabilities, while production performance depends on the ordered processors, memory population, storage backplane, network adapters, accelerators, firmware and cooling design.
HPE HPE Cray XD670 is a HPE 5U GPU-accelerated server designed for AI training, large-model inference, accelerated analytics and high-performance computing. It provides a configurable enterprise server CPU platform with enterprise ECC memory, high-speed storage, networking and management options, plus accelerator-oriented expansion for GPU-intensive workloads. Final CPU, memory, storage, accelerator and network selections should be matched to the deployment target.
Key features
What matters when evaluating HPE HPE Cray XD670
Core platform capabilities and practical deployment considerations for the named model.
Liquid-cooling option
Direct liquid cooling is available for higher-density, power-efficient deployment planning.
High-speed storage
Modern NVMe and enterprise storage options support fast datasets, databases and application tiers.
Data-center networking
Platform expansion supports high-speed network adapters for east-west traffic, storage fabrics and cluster connectivity.
Scale-out fabric
Designed to pair with high-speed Ethernet or InfiniBand-class fabrics for distributed training and accelerated clusters.
Serviceability
Rack-server design emphasizes redundant power, replaceable components, health monitoring and controlled maintenance.
Applications
Where this platform fits
Representative professional use cases. Final suitability depends on the exact configuration and application requirements.
AI training
GPU-accelerated training for language, vision and multimodal models.
Large-model inference
High-throughput serving of memory-intensive production AI workloads.
HPC & simulation
Scientific computing, engineering simulation and research workloads.
Accelerated analytics
GPU-enabled data processing, vector search and analytics pipelines.
Private AI cloud
On-premises infrastructure for controlled enterprise AI platforms.
Research clusters
Multi-node institutional, university and R&D compute environments.
Technical specifications
Product and configuration data
Use these model-level specifications for evaluation, then confirm the exact SKU, revision and ordered configuration before deployment.
- Form factor
- 5U single-node GPU-accelerated server
- Accelerators
- 8 NVIDIA H100 or H200 Tensor Core SXM5 GPUs
- Processors
- Dual-socket Intel Xeon platform; generation depends on H100/H200 configuration
- Memory
- Up to 32 DDR5 DIMMs on published platform configurations
- Cooling
- Air cooling with direct-liquid-cooling option
- System class
- 5U GPU-accelerated server
- Processor platform
- enterprise server CPU platform
- Storage
- NVMe/SAS/SATA options according to chassis and backplane configuration
- Networking
- High-speed Ethernet/InfiniBand/OCP/PCIe networking options vary by platform
- Management
- HPE enterprise remote-management and lifecycle tooling appropriate to the platform
- Power & cooling
- Redundant power and platform-specific air or liquid-cooling options; validate rack and facility requirements
- Primary workload
- AI training, large-model inference, accelerated analytics and high-performance computing
Package & configuration
Specify the complete deployment, not only the model number
The final quotation should make the exact configuration explicit so procurement, installation and acceptance teams are working from the same bill of materials.
System configuration
Server chassis with the processors, memory, storage, networking, accelerators and management options listed in the final BOM.
- Exact manufacturer and model/SKU
- Configuration and region-specific items
- Condition, packaging and included standard accessories confirmed in quotation
Deployment accessories
Rails, power cords, transceivers, rack/cooling accessories, operating-system items and support coverage are specified according to the deployment.
- Installation/deployment prerequisites
- Optional accessories and workflow components
- Support, training or service requirements where applicable
FAQ
Buyer questions
Practical checks to complete before the final purchase order.
Can HPE HPE Cray XD670 be configured to my workload?
Yes. Server CPU, memory, storage, networking, accelerators and support should be selected from the platform’s supported configuration matrix and matched to the intended workload.
What rack and facility information is needed before ordering?
Provide rack depth, available U space, power feeds, cooling capacity, network fabric, ambient conditions and any liquid-cooling requirements for accelerator-dense systems.
Can the server be supplied for AI or GPU workloads?
Where the platform supports accelerators, the complete GPU topology, firmware, power, cooling, host memory and networking must be specified as one validated configuration.
Which operating systems are supported?
Supported operating systems and hypervisors depend on platform generation, selected hardware and vendor certification. Confirm the exact OS/driver combination before deployment.
What should be included in a production quotation?
A complete BOM should identify processors, DIMMs, drives, controllers, NICs, accelerators, PSUs, rails, support level, operating-system requirements and any rack or cooling accessories.
HPEManufacturer
Manufacturer context
HPE
HPE supplies enterprise compute platforms for production data centers. The exact configuration of HPE HPE Cray XD670 should be matched to workload, rack, power, cooling, network and support requirements before deployment.
AI Robot Supplier
Procurement support for technical equipment
Use the quotation stage to lock the exact model, configuration, compatibility requirements and delivery package before shipment.
Confirm the named SKU/revision and required accessories before order release.
Match the equipment to workload, facility, software and integration requirements.
Align packing, insurance, destination requirements and receiving/site readiness.
Keep the quotation, BOM and available product documentation aligned for deployment.
Related products
Compare HPE HPE Cray XD670 with related equipment
Explore the Networking & AI Infrastructure catalogue, browse related products, or contact AI Robot Supplier for model, configuration and delivery checks. Manufacturer-level product-family information is available from HPE.
Next step
Need the exact HPE HPE Cray XD670 configuration?
Send the intended use, destination and required configuration so the quotation can be matched to the real deployment.