“Helios” architecture, sovereign AI ambitions, and Europe’s next supercomputer mark a major leap forward in global AI and HPC innovation
AMD has expanded its strategic collaboration with Hewlett Packard Enterprise (HPE) to accelerate development of open, scalable AI infrastructure tailored for hyperscale, enterprise, and sovereign computing needs. At the centre of this advancement is AMD’s new “Helios” rack-scale AI architecture, which HPE will be among the first global system providers to adopt.
“Helios” integrates AMD EPYC™ CPUs, AMD Instinct™ GPUs, Pensando advanced networking, and the ROCm™ open software stack into a unified AI platform optimized for massive cluster deployment. Alongside this, HPE is embedding a purpose-built HPE Juniper Networking scale-up switch, engineered in collaboration with Broadcom, to support Ultra Accelerator Link over Ethernet (UALoE)—delivering high-bandwidth, low-latency communication essential for frontier-scale AI.
“With Helios, we’re taking our long-standing collaboration with HPE further, bringing together the full stack of AMD compute technologies with HPE’s system innovation to deliver an open, rack-scale AI platform for the AI era.”
Dr. Lisa Su, Chair and CEO, AMD
Antonio Neri, President and CEO, HPE, said, “For more than a decade, HPE and AMD have pushed the boundaries of supercomputing. With Helios and our new scale-up networking solution, we are giving cloud providers faster deployments, greater flexibility, and reduced risk as they scale AI compute.”
The “Helios” platform delivers up to 2.9 exaFLOPS of FP4 performance per rack, built on the OCP Open Rack Wide standard. This allows customers to deploy AI infrastructure faster, streamline power and cooling requirements, and scale workloads from enterprise AI to advanced HPC research.
A major outcome of this collaboration is Herder, Europe’s next-generation supercomputer for the High-Performance Computing Center Stuttgart (HLRS). Built on the HPE Cray GX5000 platform and powered by AMD Instinct MI430X GPUs and next-generation AMD EPYC “Venice” CPUs, Herder will support both traditional numerical simulation and cutting-edge AI workloads.
Dr. Michael Resch, Director, HLRS, added, “Herder’s architecture lets us support HPC and AI simultaneously, enabling larger simulations, hybrid workflows, and entirely new computational methods for scientific discovery.”
Scheduled for deployment in late 2027, Herder will replace HLRS’s current flagship system and significantly advance Europe’s sovereign AI and HPC capabilities.