HPE ProLiant Compute XD685 (AMD MI300X) Server

The HPE ProLiant Compute XD685 (part number S2D60A) is a purpose-built, high-density GPU compute node engineered specifically for large-scale generative AI training and inference workloads. Based on AMD Instinct MI300X accelerators — which integrate both GPU compute dies and HBM3 memory in a unified architecture — the XD685 delivers exceptional memory bandwidth and on-package memory capacity, enabling models with hundreds of billions of parameters to be trained or served with reduced inter-node data movement. The node is designed to operate within HPE's XD chassis-based infrastructure, providing a dense, thermally optimized platform suited to modern AI data center deployments.

Each XD685 compute node is equipped with multiple AMD Instinct MI300X APUs interconnected via AMD Infinity Fabric, enabling high-bandwidth GPU-to-GPU communication within the node. The MI300X accelerator's unified memory architecture — combining GPU compute with up to 192 GB of HBM3 memory per accelerator — dramatically increases the addressable memory pool available to large language models and multimodal AI frameworks without requiring model parallelism across additional nodes. Host connectivity is provided by AMD EPYC processors, and the platform supports high-speed networking fabrics such as InfiniBand and high-bandwidth Ethernet to integrate into GPU compute clusters at scale.

The ProLiant Compute XD685 is a central element of HPE's AI supercomputing portfolio, validated for deployment within HPE Cray XD systems and compatible with leading AI software stacks including ROCm, PyTorch, and JAX. Its architecture targets enterprise customers, national laboratories, and cloud service providers building dedicated generative AI infrastructure, LLM fine-tuning pipelines, and large-scale AI inferencing services where memory capacity per accelerator and aggregate cluster bandwidth are primary design constraints.

Ideal for

• Large language model (LLM) pre-training and fine-tuning at scale, leveraging the high HBM3 memory capacity of AMD Instinct MI300X to fit models with hundreds of billions of parameters on fewer nodes
• Generative AI inference serving for enterprise applications requiring low-latency, high-throughput responses from very large foundation models
• Multimodal AI workloads combining vision, language, and structured data, where unified high-bandwidth memory reduces data staging bottlenecks
• High-performance computing (HPC) simulations — including molecular dynamics, climate modeling, and computational chemistry — that benefit from the MI300X unified CPU-GPU memory architecture
• AI model parallelism and pipeline parallelism research within GPU clusters connected via InfiniBand or high-bandwidth Ethernet fabrics
• Enterprise AI infrastructure consolidation, replacing heterogeneous GPU fleets with a standardized, HPE-managed high-density compute platform integrated into HPE GreenLake or on-premises data center operations

Technical specifications

Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.