HPE Cray GX5000 with 6th gen AMD EPYC targets dense AI/HPC racks

HPE highlighted the HPE Cray Supercomputing GX5000 platform configured with 6th Gen AMD EPYC processors for AI and HPC workloads at AMD Advancing AI 2026 in San Francisco. The system is aimed at high-density deployments, and HPE tied the update to both performance-per-rack and facility-efficiency claims.

The HPE Cray Supercomputing GX5000 is positioned as a fully integrated supercomputing system spanning CPUs, accelerators, storage, networking, software, and 100% direct-liquid cooling. HPE says the GX5000 delivers “40% more AMD EPYC 9996 cores per rack than the market’s next densest offering.” On the facilities side, HPE says the platform needs 25% less floor space than the previous generation and can increase warm facility water temperature by 25%, with up to 40 C heating capacity, enabled by next-generation direct-liquid cooling.

Density and heat reuse are increasingly tied together in modern supercomputing builds: packing more compute into fewer racks shifts the constraint to power delivery and liquid-side thermal design. Claims like “40% more cores per rack” and warmer water operation are useful engineering targets, but they’re only meaningful once you map them to your own node mix, redundancy model, and CDU and facility-water constraints.

Blade options and per-rack figures

HPE described two blade configurations for the GX5000 with 6th Gen AMD EPYC processors:

The HPE Cray Supercomputing GX350a Accelerated Blade combines one 6th Gen AMD EPYC processor with four AMD Instinct MI430X GPUs, with support for up to 112 AMD GPUs per rack.

The HPE Cray Supercomputing GX250 Compute Blade is a CPU-only option for double-precision workloads, with eight 6th Gen AMD EPYC processors per node. HPE says a single rack can house 40 nodes for up to 81,920 CPU cores.

Early adopters: ORNL and HLRS

HPE named Oak Ridge National Laboratory (ORNL) and the High-Performance Computing Center of the University of Stuttgart (HLRS) as early adopters of GX5000 systems with 6th Gen AMD EPYC processors. ORNL, HPE, and AMD are working on an exascale system called “Discovery,” described as an HPE Cray Supercomputing GX5000 system with new AMD EPYC processors. HPE said Discovery will support converged simulation and AI workloads, AI model development and training, and act as a testbed for ORNL’s quantum-computing efforts.

Separately, HPE said ORNL, HPE, and AMD are building a dedicated AI system called “Lux” using next-generation HPE systems with 6th Gen AMD EPYC processors, intended to accelerate AI development as part of the US Genesis Mission.

In Germany, HPE said HLRS’ upcoming flagship system, “Herder,” will be a GX5000 system with new AMD EPYC CPUs and is expected to be 7X more powerful than HLRS’ current supercomputer.

AMD Helios AI rack scale system with UALoE

HPE also said it is collaborating with AMD on the AMD Helios AI rack scale system. HPE described Helios as delivering 260 TB/s of scale-up bandwidth and 2.9 exaflops of FP4 performance, and said the networking uses purpose-built HPE Juniper Networking hardware and software based on the Ultra Accelerator Link over Ethernet (UALoE) standard.

More information on HPE Cray Supercomputing is available at HPE’s Cray exascale supercomputing page.