HPE Unveils Next-Gen Cray Supercomputer Line

HPE introduces three multi-partner, multi-workload compute blades for its next-generation supercomputing platform achieving one of the highest compute performance densities in the industry
The HPE Supercomputing Management Software delivers a secure systems management experience for AI and HPC infrastructure across all phases of the system life cycle
University of Stuttgart and Leibniz Supercomputing Centre select the new HPE Cray Supercomputing GX5000 platform, validating industry leadership innovation in the new solutions

Hewlett Packard Enterprise (HPE) has announced significant additions to its next-generation HPE Cray supercomputing portfolio. The expansion focuses on achieving industry-leading compute density to meet the growing demands of both high-performance computing (HPC) and artificial intelligence (AI). This new architecture is designed to support high-scale productivity for research laboratories, sovereign entities, and large enterprises. The updates include new multi-partner, multi-workload compute blades, streamlined management software, and enhanced high-performance interconnectivity. This comprehensive approach establishes a powerful foundation for scientific discovery and advanced simulation.

Trish Damkroger, HPE’s Senior Vice President and General Manager for HPC and AI Infrastructure Solutions, noted that global organizations require improved computing performance across all their workloads. She highlighted the platform as the answer to customers seeking higher performance density within a unified AI and HPC framework. HPE views supercomputing and AI as essential engines for innovation that will advance how people live and work. The company previously introduced the HPE Cray Supercomputing GX5000 and the HPE Cray Supercomputing Storage Systems K3000, which features embedded open-source Distributed Asynchronous Object Storage (DAOS) software for increased performance.

Accelerated Industry Adoption

The new HPE Cray Supercomputing platform is already seeing rapid adoption within the HPC community. The High-Performance Computing Center of the University of Stuttgart (HLRS) and the Leibniz Supercomputing Centre (LRZ) have both selected the HPE Cray Supercomputing GX5000 for their upcoming systems. These include the Herder supercomputer at HLRS and the Blue Lion supercomputer at LRZ, both poised to unlock new scientific horizons for their users. Professor Dr.-Ing. Michael Resch, Director at HLRS, praised HPE’s systems for consistently enhancing the ability of their user communities to achieve novel discoveries and design better technologies.

Professor Dr. Dieter Kranzlmüller, Chairman of the Board of Directors at LRZ, emphasized the commitment to efficiency and sustainability in their choice. The Blue Lion system will feature 100 percent direct liquid cooling capable of operating at temperatures up to $40^\circ\text{C}$, enabling the reuse of waste heat across the Garching research campus. This flagship system is expected to deliver sustained performance up to 30 times faster than the current one, allowing researchers to seamlessly integrate traditional modeling with cutting-edge AI methodologies. The strong initial uptake demonstrates confidence in the platform’s capabilities for supporting complex, converged workloads.

New High-Density Compute Blades

Three new processing blades are at the core of this portfolio expansion, all utilizing 100 percent direct liquid cooling to maximize performance density. These blades support flagship CPUs and GPUs, including the next-generation NVIDIA Rubin platform, AMD Instinct™ MI430X, and the next-gen AMD EPYC™ processors, codenamed “Venice.” Each blade offers a choice of either four or eight HPE Slingshot 400 gigabits per second (Gbps) endpoints and an option for two non-volatile memory express (NVMe) solid-state drives (SSDs). Customers can mix and match the three types of blades within a single HPE Cray Supercomputing GX5000 compute rack to configure the optimal setup for their specific workloads.

The HPE Cray Supercomputing GX440n Accelerated Blade provides a universal compute engine for mixed-precision computing, featuring four NVIDIA Vera CPUs and eight NVIDIA Rubin GPUs, achieving industry-leading GPU density of up to 192 NVIDIA Rubin GPUs per rack. Alternatively, the HPE Cray Supercomputing GX350a Accelerated Blade combines one next-generation AMD EPYC “Venice” CPU with four AMD Instinct™ MI430X GPUs, a new accelerator specifically engineered for sovereign AI and HPC. Finally, the HPE Cray Supercomputing GX250 Compute Blade is designed for CPU-only partitions for double-precision workloads, packing eight next-generation AMD EPYC “Venice” CPUs, which delivers industry-leading flagship x86 CPU core density.

Unified Management and Connectivity

HPE is also enhancing the operational side with the new HPE Supercomputing Management Software. This software introduces capabilities to support multi-tenant, virtualized, and containerized environments, offering customers greater flexibility to deploy converged AI and HPC processing while enabling workload and user group isolation. Crucially, the software provides system-wide power and energy management, allowing customers to monitor usage, estimate consumption, and integrate with power-aware schedulers to maximize energy efficiency and better anticipate costs. This system delivers a unified and secure management experience across the supercomputer’s entire life cycle, including provisioning, monitoring, power, cooling, and scaling, complete with strengthened security features.

The HPE Slingshot 400 interconnect is now available for the HPE Cray Supercomputing GX5000 platform, engineered for the denser form factor and designed to perform under heavier converged AI and HPC workloads. A new chassis was developed for the 100 percent direct liquid-cooled switch blade, featuring 64 $400\text{ Gbps}$ ports. The Slingshot 400 switches leverage the higher performance topology of the GX5000, offering customers reduced latency, improved sustained bandwidth, and reliability while effectively managing system costs. The new supercomputing GX440n and GX250 compute blades, as well as the management software and Slingshot 400 availability, are expected in early 2027, while the K3000 storage system will be available earlier, in early 2026.