Overview
- HPE will be one of the first system providers to adopt Helios, an AMD rack-scale architecture built around EPYC CPUs, Instinct GPUs, Pensando networking and the ROCm software stack.
- AMD cites up to 2.9 exaFLOPS of FP4 performance per rack using Instinct MI455X GPUs and next-generation EPYC “Venice” CPUs, positioning Helios for large AI and HPC workloads.
- HPE plans worldwide availability of Helios-based systems in 2026, aligning the design with OCP Open Rack Wide standards to speed deployment.
- Networking is anchored by a purpose-built HPE Juniper scale-up switch developed with Broadcom that uses the UALoE standard to deliver high bandwidth and low latency across large clusters.
- The partners also announced Herder for Germany’s HLRS, an HPE Cray GX5000 system using AMD Instinct MI430X GPUs and EPYC “Venice” CPUs, with delivery slated for the second half of 2027 and service targeted by year-end 2027.