HPE Servers (ProLiant) · Server Hardware

HPE ProLiant Compute XD (Apollo) — AI/HPC

Name: HPE ProLiant Compute XD (Apollo) — AI/HPC — HPE Servers (ProLiant)
Brand: HPE Servers (ProLiant)
Availability: InStock

AI training and HPC platform with up to 8 GPUs.

Request a quote FAQ

Quick answer

HPE ProLiant Compute XD (formerly Apollo) is a high-density AI training and HPC platform supporting up to 8 NVIDIA H100 / H200 GPUs. It is HPE's enterprise alternative to NVIDIA DGX systems for large-language-model training and HPC workloads.

HPE Compute XD685 is a 6U server designed for AI training. It hosts 8 NVIDIA H100 or H200 SXM GPUs interconnected via NVLink and NVSwitch — the same HGX board used by NVIDIA DGX H100. HPE GreenLake enables a subscription-based consumption model as an alternative to upfront capital purchase.

InfiniBand 400G networking, liquid-cooling options and enterprise-grade serviceability make the platform a credible HPE alternative to NVIDIA DGX for enterprise AI infrastructure across the EU, UK and US.

Key features

What it offers

Up to 8 NVIDIA H100/H200 GPUs
NVLink and NVSwitch topology
Liquid-cooling option
InfiniBand 400G networking
HPE GreenLake subscription model option

Tech Summary

Important technical data

GPU: 8x NVIDIA H100/H200 SXM5
Topology: NVLink + NVSwitch
Networking: InfiniBand 400G
Cooling: Air or liquid

Use Cases

When would you choose this product?

LLM training

A major EU technology company's LLM pre-training cluster.

University research

Research HPC

University AI research laboratory.

Healthcare

Medical-imaging AI

Hospital AI-driven diagnostic model training.

Who is it for?

AI training, LLM and HPC customers across EU, UK and US.

Frequently Asked Questions

Frequently asked questions

How does Compute XD685 compare to Dell PowerEdge XE9680?

Very similar — both use the same NVIDIA HGX H100 baseboard. The deciding factor is management experience (HPE iLO + OneView versus Dell iDRAC) and regional service network.

Is it a NVIDIA DGX alternative?

Yes. Same HGX H100 baseboard, more flexible HPE service options and typically a better commercial proposition.

Is liquid cooling required?

Air cooling is available, but with 8x H100 the thermal load is high; liquid cooling improves data-center PUE and is increasingly the default for new builds.

Why InfiniBand 400G?

Multi-node training requires low-latency, high-bandwidth interconnect for tensor-parallel and pipeline-parallel distributed training. InfiniBand 400G is the current enterprise standard.

What is HPE GreenLake?

GreenLake is HPE's CAPEX-to-OPEX consumption model — a subscription-based, pay-per-use offering for AI infrastructure.

What is the GPU lead time?

H100 and H200 supply remains constrained; typical order-to-delivery is 8-16 weeks. Sora Yazılım plans GPU procurement well in advance against customer pipeline.

Warranty?

Three years standard with optional Tech Care 24/7 4-hour SLA upgrade.

Vendor's official product page

Opens the vendor's original technical documentation and product page in a new tab.

HPE Servers (ProLiant) — HPE ProLiant Compute XD (Apollo) — AI/HPC →

From the same brand

Comparable Solutions

Services we deliver alongside this product

yapay-zeka-ai-llm