Hewlett Packard Enterprise (HPE) has unveiled the HPE ProLiant Compute XD685, a server designed to enhance the training of sophisticated and large AI models. In partnership with AMD, the XD685 utilizes AMD Instinct MI325X accelerators and 5th Generation AMD EPYC processors to provide a high-performance, energy-efficient AI infrastructure.

This new server system is aimed at streamlining the quick deployment of AI clusters for multiple applications, such as multi-modal training, natural language processing, and the development of extensive language models.

As the urgency to leverage AI capabilities escalates, the need for solutions that can deliver quicker results in fields like healthcare, climate science, and productivity has increased. Governments, significant AI developers, and service providers require flexible, swiftly deployable infrastructure to fully harness the advantages of AI technology. The launch of the HPE ProLiant Compute XD685 seeks to fulfill these requirements, providing businesses with the necessary tools to accelerate AI integration and achieve notable advancements.

Trish Damkroger, HPE’s Senior Vice President and General Manager for HPC and AI Infrastructure Solutions, noted the pivotal role of the system’s design in facilitating large-scale AI training. “To train large language models efficiently, substantial parallel computing power and specialized services are essential, which HPE’s high-performance computing solutions can offer,” commented Trish Damkroger. “With the HPE ProLiant Compute XD685, our collaboration with AMD is enhanced to support AI model creators and propel scientific and engineering achievements.”

Forrest Norrod, Executive Vice President and General Manager of AMD’s Data Center Solutions Group, highlighted the transformative nature of combining AMD’s latest processors and accelerators for AI performance. “The HPE ProLiant Compute XD685 provides clients with a significant edge in expanding AI initiatives while boosting efficiency and minimizing market introduction times,” he stated.

Air and Direct Liquid Cooling

The HPE ProLiant Compute XD685 boasts a new modular chassis designed to improve speed and flexibility for substantial AI model training operations. The 5U chassis facilitates various configurations, allowing for different GPUs, CPUs, and cooling approaches. This modularity enables companies to minimize time to market by accommodating diverse components and software configurations.

Engineered with options for both air and direct liquid cooling, the XD685 reflects HPE’s long-standing specialization in liquid-cooled IT systems, delivering a sustainable and cost-efficient answer for AI infrastructure. Direct liquid cooling is particularly advantageous for organizations striving to meet increasing power demands while advancing sustainability goals and reducing operational costs.

The server’s architecture enhances AI workloads by accommodating up to eight AMD Instinct MI325X or MI300X accelerators, along with two AMD EPYC 9005 Series CPUs in a compact 8-nodes-per-rack layout. The AMD Instinct MI325X accelerators, built on the CDNA 3 architecture, provide peak performance with 6 TB/s memory bandwidth and HBM3E memory, allowing enterprises to train large AI models with fewer GPUs and enhanced cost-effectiveness.

Security remains an essential priority for the HPE ProLiant Compute XD685, which employs HPE iLO technology for secure system management. This solution guarantees a secure supply chain starting at the silicon level and includes integrated management features to protect against escalating security risks. The deployment of HPE iLO facilitates streamlined administration and strong security protocols, addressing the evolving challenges within AI and data-intensive computing settings.

Management Processes Bare Metal

To simplify the management of extensive AI clusters, the system incorporates HPE Performance Cluster Manager, a comprehensive management tool that streamlines complex tasks such as system setup, GPU stress testing, and ongoing cluster health assessments. This tool automates management functions from the ground up, ensuring AI clusters function efficiently and effectively.

The modular construction and built-in cooling systems enable the XD685 to cater to a variety of enterprise requirements, whether through HPE’s specialized deployment services or the server’s inherent adaptability to accommodate various configurations. HPE’s services offer customized solutions for AI cluster implementations, including on-site deployment and thorough factory testing to guarantee stability and optimal functionality.

In conjunction with the introduction of the XD685, HPE announced that its ProLiant Gen11 servers equipped with AMD EPYC 9005 Series processors have set 48 world records across multiple performance standards, including power efficiency for business workloads. These achievements reinforce HPE’s dominance in high-performance computing and highlight the capabilities of AMD’s latest processors.

The HPE ProLiant Compute XD685 is set to be available for purchase in the first quarter of 2025. With its robust capabilities and energy-efficient design, the XD685 is positioned to be a leading solution for organizations seeking to enhance AI-driven innovation and address the escalating requirements of AI model training.

As AI technology continues to evolve, HPE‘s partnership with AMD aims to provide flexible, high-performance solutions that streamline AI training and spur competitive innovation across sectors. With the launch of the XD685, HPE is well-equipped to support a broad range of AI applications.