HPE unveils ProLiant XD685 for advanced AI model training

Hewlett Packard Enterprise has introduced the HPE ProLiant Compute XD685, a new solution aimed at enhancing the training of large, complex AI models leveraging AMD's advanced technology.

The new system is equipped with 5th Gen AMD EPYC processors and AMD Instinct MI325X accelerators, designed to facilitate high-performance AI clusters. This offering targets large language model training, natural language processing, and multi-modal training.

According to the company, AI service providers, government entities, and large model builders are in need of high-performance solutions capable of quick deployment. "Training large language models, and doing so efficiently, requires strong-scaling, massive parallel computing capabilities, and unique services that only HPE's high performance computing solutions deliver," said Trish Damkroger, Senior Vice President and General Manager, HPC & AI Infrastructure Solutions at HPE. "As a leader in delivering the world's most powerful and energy-efficient systems, together with AMD, we are extending this innovation to the new HPE ProLiant Compute XD685 to support a growing market of AI model builders and accelerate scientific and engineering breakthroughs across industries."

Forrest Norrod, Executive Vice President and General Manager, Data Centre Solutions Group, AMD, commented on the collaboration: "Through the powerful combination of the latest AMD EPYC processors and Instinct accelerators, HPE ProLiant Compute XD685 customers can accelerate time to market and experience true leadership in AI performance and efficiency. Our collaboration with HPE continues to deliver flexible, highly performant solutions that help maximise AI efforts for increased demand in large language model training and advances competitive industry innovation."

HPE's engineering advancements include a new modular 5U chassis for the ProLiant Compute XD685, designed to expedite solution delivery times while providing flexibility for varied hardware and software configurations. The system integrates several features, including options for air and direct liquid cooling to enhance energy efficiency and HPE iLO for secure management.

The system boasts the capacity to run large AI models with fewer GPUs thanks to its AMD CDNA 3 architecture, featuring AMD Instinct MI325X accelerators. These accelerators offer HBM3e memory capacity and an impressive memory bandwidth of 6 TB/s, enhancing performance and reducing total cost of ownership (TCO).

The direct liquid cooling option allows organisations to manage rising power needs, support sustainability initiatives, and lower operational expenses, drawing on HPE's extensive experience with liquid-cooled IT environments.

The HPE ProLiant Compute XD685 also benefits from HPE Services, which provides a comprehensive suite of services for setting up and operating large AI clusters globally. This includes configuration, validation, and testing to minimise time-to-deployment.

Security has been underscored with HPE iLO, offering an advanced security management protocol starting at the production level and rooted into the silicon.

The ProLiant Compute XD685 maximises rack density with its compact 8-node-per-rack configuration, supporting either the MI325X or MI300X accelerators alongside the latest AMD EPYC 9005 Series processors.

HPE's Performance Cluster Manager simplifies the management of extensive AI clusters, automating system setup and maintaining operational health through detailed telemetry and testing.

Additionally, HPE announced 30 new world records set on its ProLiant Gen11 servers, featuring the AMD EPYC 9005 Series processors. These records highlight leading performance and power efficiency in enterprise workloads.

The HPE ProLiant Compute XD685 is now available for order, with general availability expected in the first quarter of 2025 as configurations are gradually introduced.