HPE AI Servers
Purpose-built for AI training, tuning and inferencing workloads of any size, with the right combination of performance and scalability.
Industry-leading AI performance
HPE secures 18 #1 rankings in MLPerf® Inference v6.0 benchmarks.
Optimized performance at any scale
Engineered for AI training, tuning and inference. Delivering optimized AI performance from edge to core to thousands of racks.
Innovation for secure, efficient AI
From built-in security and lifecycle protection across your entire infrastructure, to advanced liquid cooling technology, HPE innovation gives you the confidence to accelerate AI outcomes.
AI expertise from design to operations
Accelerate deployment and achieve consistent operational stability worldwide with expert guidance from HPE AI Services.
Bring AI to your business everywhere
Purpose-built portfolio of HPE Servers for AI workloads.
- Servers for large-scale training, tuning and inferencing
- Servers for AI inferencing
 HPE ProLiant Compute XD685 |  HPE Compute XD690 | |
|---|---|---|
| Description | Ideal to accelerate large AI model training sustainably and securely, and powered by a choice of eight NVIDIA or AMD Instinct™ GPUs. | Designed for the escalating requirements of large AI model training, tuning and inferencing. |
| GPU | 8x NVIDIA H200 (air or DLC) | 8 X NVIDIA B300 air-cooled |
| CPU | 2x 5th Generation AMD EPYC™ Processors | 2x Intel® Xeon® 6 processors |
| Form factor | 5U DLC, 6U air-cooled | 10U |
| Cooling | Air or liquid, depending on GPU | Air cooled |
| Management | Server: HPE iLO6 | Server: BMC, Redfish APIs, 1GbE LAN |
 ProLiant DL145 |  ProLiant EL2000 |  ProLiant ML350 |  ProLiant DL345 |  ProLiant DL365 |  ProLiant DL380 |  ProLiant DL385 |  ProLiant DL380a | |
|---|---|---|---|---|---|---|---|---|
| GPU | NVIDIA RTX PRO™ 4500 Blackwell Server Edition | NVIDIA RTX PRO 4500 & 6000 Blackwell | NVIDIA RTX PRO 4500 Blackwell | NVIDIA RTX PRO 4500 & 6000 Blackwell | NVIDIA RTX PRO 4500 Blackwell | NVIDIA RTX PRO 4500 & 6000 Blackwell | NVIDIA RTX PRO 4500 & 6000 Blackwell | NVIDIA RTX PRO™ 6000 Blackwell Server Edition |
| CPU | 1P AMD EPYC 8005 | 1P INTEL Xeon 6 | 2P INTEL Xeon 6 | 1P 5th Gen AMD EPYC | 2P 5th Gen AMD EPYC | 2P INTEL Xeon 6 | 2P 5th Gen AMD EPYC | 2P INTEL Xeon 6 |
| Form factor | 2U Compact | 2U SWaP optimized | Performance Tower | 2U Rack optimized | 1U Rack optimized | 2U Rack optimized | 2U Rack optimized | 4U Rack optimized |
| Cooling | Air | Air | Air | Air | Air, DLC | Air | Air, DLC | Air, DLC |
| Max memory | Up to 768GB | Up to 2TB | Up to 8TB | Up to 6TB | Up to 6TB | Up to 8TB | Up to 6TB | Up to 8TB |
| Storage | SFF or EDSFF | M.2 + EDSFF | SFF, EDSFF, LFF drives | SFF, EDSFF, LFF drives | SFF, EDSFF drives | SFF, EDSFF, LFF drives | SFF, EDSFF, LFF drives | SFF, EDSFF drives |
| Management | HPE iLO 6 , Compute Ops Management | HPE iLO 7 , Compute Ops Management | HPE iLO 7 , Compute Ops Management | HPE iLO 7 , Compute Ops Management | HPE iLO 7 , Compute Ops Management | HPE iLO 7 , Compute Ops Management | HPE iLO 7 , Compute Ops Management | HPE iLO 7 , Compute Ops Management |
Our customers
Bring AI to your enterprise everywhere—edge to core
HPE ProLiant Compute and NVIDIA RTX PRO Everywhere
Enterprises need a unified approach that delivers accelerated performance, simplifies deployment, reduces cost, and raises security standards—all without redesigning their data centers. See how HPE and NVIDIA unleash universal acceleration from edge to data center.
Take the next steps
Ready to get started? Explore purchasing options or engage with HPE experts to determine the best solution for your business needs.
Our partners
Related Products
News and resources
HPE Servers for large-scale AI training, tuning and inferencing FAQs
What is the portfolio of HPE Servers for large-scale AI training, tuning and inferencing?
It is a purpose-built portfolio of high-performance HPE servers designed to help service providers, model builders, large enterprises and sovereign organizations train, tune, and run inference on large AI models at scale. The portfolio includes HPE ProLiant Compute XD685, HPE Compute XD690, and HPE Compute XD700, bringing together GPU density, scalable management, advanced cooling, and HPE services expertise.
What workloads are these HPE servers designed to support?
These systems are engineered for demanding AI workloads such as large language model training, model fine-tuning, natural language processing, multimodal training, AI reasoning, and high-throughput inference.
How do HPE ProLiant Compute XD685, HPE Compute XD690, and HPE Compute XD700 fit together?
Each system addresses large-scale AI needs with different choices of accelerator technology, cooling approach, and deployment profile. HPE ProLiant Compute XD685 offers flexible NVIDIA or AMD accelerator options, HPE Compute XD690 supports NVIDIA Blackwell Ultra GPUs in an air-cooled design, and HPE Compute XD700 will support the latest NVIDIA Rubin GPUs.
What makes HPE differentiated for large-scale AI infrastructure?
HPE combines purpose-built AI servers with deep experience delivering some of the world's largest and most complex high-performance computing environments. Customers can also benefit from HPE innovations in direct liquid cooling, secure management, global supply chain execution, performance engineering, and end-to-end services that help accelerate deployment and stabilize operations.
What customer value does this portfolio deliver?
The portfolio helps organizations shorten time-to-results, scale AI projects with confidence, and improve operational efficiency as model size and workload complexity grow. By combining accelerated servers with HPE deployment, support, and lifecycle services, customers can move from pilots to production with less infrastructure risk.
How does HPE help customers address power and cooling challenges for AI?
Large AI models place intense demands on data center power and thermal capacity. HPE helps address these challenges with systems that support air cooling or direct liquid cooling, depending on the platform and configuration, helping customers improve energy efficiency, support higher-density deployments, and plan for the next generation of accelerated AI infrastructure. With HPE's hundreds of patents and extensive experience spanning five decades in the liquid cooling space, customers can have peace of mind as they implement large liquid-cooled AI environments.
Can this portfolio of servers be used for both training and inference?
Yes. While these servers are primarily designed for large-scale training and tuning, they can also support high-performance inference for large and complex models. This flexibility helps organizations use the same AI infrastructure strategy across model development, optimization, and production deployment.
What are examples of how customers can use this portfolio?
Customers can use these systems to build AI factories, train domain-specific language models, accelerate computer vision and multimodal AI, support sovereign AI initiatives, or deliver cloud-like AI services. Specific examples include Subaru using HPE Cray XD670 servers to advance next-generation driver assist systems and Oakridge National Laboratory planning a multi-tenant AI platform with the Lux AI cluster, based on HPE ProLiant Compute XD685.
How is this portfolio positioned for future AI requirements?
As AI models become larger, more agentic, and more inference-intensive, infrastructure must deliver greater performance, efficiency, and scalability. HPE is evolving this portfolio with newer accelerator technologies, advanced cooling options, and global services designed to help customers scale from individual systems to large AI clusters and AI factories.