AMD, Broadcom, and HPE announced their collaboration to accelerate the next generation of open, scalable AI infrastructure powered by AMD computing technologies. HPE will be among the first system providers to adopt AMD’s “Helios” rack-scale AI architecture. This architecture will include a specially designed HPE Juniper Networking scale-up switch and software for easy, high-bandwidth Ethernet connectivity.
Helios Architecture
“Helios” combines AMD EPYC CPUs, AMD Instinct GPUs, AMD Pensando networking, and the AMD ROCm open software stack. This creates a unified platform that emphasizes performance, efficiency, and scalability. The system aims to simplify the setup of large-scale AI clusters, allowing for faster solutions and more adaptable infrastructure across research, cloud, and enterprise settings.
The AMD “Helios” rack-scale AI platform can deliver up to 2.9 exaFLOPS of FP4 performance per rack using AMD Instinct MI455X GPUs, next-generation AMD EPYC “Venice” CPUs, and AMD Pensando Vulcano NICs for scale-out networking. All of this is connected through the open ROCm software ecosystem, which allows for flexibility and innovation across AI and HPC workloads.
OCP Open Rack Design
Built on the OCP Open Rack Wide design, “Helios” helps customers and partners speed up deployment and provides a flexible solution for demanding AI workloads. This has enabled HPE to incorporate unique technologies for its customers, especially a scale-up Ethernet switch and software made specifically for “Helios.” Developed in partnership with Broadcom, the switch offers enhanced AI performance through the Ultra Accelerator Link over Ethernet (UALoE) standard.
Powering Herder Supercomputer
AMD is also powering Herder, a new supercomputer for the High-Performance Computing Center Stuttgart (HLRS) in Germany. It features AMD Instinct MI430X GPUs and next-generation AMD EPYC CPUs, delivering top performance and efficiency. Built on the HPE Cray Supercomputing GX5000 platform, Herder will provide excellent performance and efficiency for HPC and AI workloads at scale.
Herder is designed to promote scientific discovery and advance Sovereign AI projects across Europe. It will support complex simulations and AI model training within a fully open, trusted infrastructure. By combining AMD’s leading computing portfolio with HPE’s reliable system design, Herder will be a powerful tool for scientific research and industrial innovation for European researchers and businesses.
Amazon