AMD today announced an expanded collaboration with Hewlett Packard Enterprise (HPE) to accelerate the development of open, scalable AI infrastructure for next-generation workloads. The partnership brings together AMD’s full compute technology portfolio within the new “Helios” rack-scale AI architecture, combined with HPE’s system design and networking innovations.
Key Announcements
- Early OEM Adoption: HPE will be one of the first system providers to adopt the AMD “Helios” architecture, an open, full-stack AI platform designed for large-scale AI clusters. Global availability of the platform through HPE is expected in 2026.
- Advanced AI Networking: “Helios” will integrate purpose-built HPE Juniper Networking switches, developed with Broadcom, to deliver high-bandwidth, low-latency connectivity optimized for massive AI deployments.
- European Supercomputing Project: AMD and HPE will jointly power “Herder,” a new supercomputer for the High-Performance Computing Center Stuttgart (HLRS) in Germany. Built on the HPE Cray GX5000 platform with AMD Instinct™ MI430X GPUs and next-generation AMD EPYC™ “Venice” CPUs, Herder will support HPC and Sovereign AI research across Europe.
Delivery is planned for the second half of 2027, with operation expected by the end of 2027.
AMD “Helios” Architecture
The AMD “Helios” rack-scale AI platform integrates AMD EPYC™ CPUs, AMD Instinct™ GPUs, AMD Pensando™ networking, and the open ROCm™ software ecosystem into a unified solution optimized for performance and efficiency.
Key capabilities include:
- Up to 2.9 exaFLOPS of FP4 performance per rack using AMD Instinct MI455X GPUs
- Next-generation AMD EPYC “Venice” CPUs
- AMD Pensando Vulcano NICs for large-scale networking
- Support for OCP Open Rack Wide design standards
- Enhanced interoperability and flexibility for cloud, enterprise, and research environments
The platform also incorporates a purpose-built HPE scale-up Ethernet switch, developed with Broadcom, featuring Ultra Accelerator Link over Ethernet (UALoE) to deliver optimized throughput for demanding AI workloads.
Herder: Europe’s Next AI-Ready Supercomputer
The Herder supercomputer will replace HLRS’s current flagship system, “Hunter,” and is designed to support traditional HPC workloads alongside advanced AI and hybrid workflows.
Prof. Dr. Michael Resch, Director of HLRS, highlighted the system’s versatility:
“Herder’s architecture enables our users to run larger simulations, pursue machine learning applications, and develop new hybrid HPC/AI methods made possible by these next-generation technologies.”
