HPE has expanded its next-generation Cray supercomputing portfolio with new compute blades, updated management software and enhanced interconnect options designed to support increasingly converged AI and high-performance computing (HPC) workloads.
Announced on 14 November, the additions include three liquid-cooled processing blades that support upcoming GPU and CPU platforms from NVIDIA and AMD. The blades are designed to deliver high compute density within the HPE Cray Supercomputing GX5000 architecture and allow mixed deployments of accelerated and CPU-only nodes in the same rack.
The release follows last month’s introduction of the GX5000 platform and the K3000 storage system, which integrates the open-source Distributed Asynchronous Object Storage (DAOS) software. The University of Stuttgart’s High-Performance Computing Center and the Leibniz Supercomputing Centre have already selected the GX5000 for their next national research systems, planned to replace existing infrastructure with Herder and Blue Lion supercomputers.
The new blades support a range of processor combinations, including NVIDIA’s Rubin GPUs, AMD Instinct MI430X accelerators and next-generation AMD EPYC CPUs. Depending on configuration, each rack can host up to 192 NVIDIA Rubin GPUs, 112 AMD MI430X GPUs or high-density CPU-only nodes aimed at double-precision workloads. The systems are designed for direct liquid cooling to support higher thermal loads.
HPE has also introduced updated Supercomputing Management Software that adds support for multi-tenant, containerised and virtualised environments. New energy and power-management tools allow operators to monitor consumption and integrate with power-aware scheduling systems, reflecting growing attention to energy efficiency in large research and enterprise environments.
The HPE Slingshot 400 interconnect, now available for GX5000-based systems, provides up to 400 Gbps switching and is engineered for denser system topologies. New switch chassis options allow scaling from 512 to 2,048 ports.
For storage, the HPE Cray K3000 system — available in 2026 — will offer multiple NVMe configurations aimed at supporting data-intensive AI workflows, with flexible drive sizes and support for Slingshot 200/400, InfiniBand NDR and 400 Gb Ethernet networking.
The new compute blades and management software are scheduled for availability in early 2027, while updated interconnects for GX5000 clusters will also arrive the same year.
The announcements come as research institutions and national computing centres continue to assess infrastructure capable of supporting both traditional modelling workloads and increasingly large-scale AI applications.
