AMD, Broadcom, and HPE introduced their collaboration to speed up the following technology of open, scalable AI infrastructure powered by AMD computing applied sciences. HPE might be among the many first system suppliers to undertake AMD’s “Helios” rack-scale AI structure. This structure will embrace a specifically designed HPE Juniper Networking scale-up change and software program for straightforward, high-bandwidth Ethernet connectivity.
Helios Structure
“Helios” combines AMD EPYC CPUs, AMD Intuition GPUs, AMD Pensando networking, and the AMD ROCm open software program stack. This creates a unified platform that emphasizes efficiency, effectivity, and scalability. The system goals to simplify the setup of large-scale AI clusters, permitting for quicker options and extra adaptable infrastructure throughout analysis, cloud, and enterprise settings.
The AMD “Helios” rack-scale AI platform can ship as much as 2.9 exaFLOPS of FP4 efficiency per rack utilizing AMD Intuition MI455X GPUs, next-generation AMD EPYC “Venice” CPUs, and AMD Pensando Vulcano NICs for scale-out networking. All of that is linked by means of the open ROCm software program ecosystem, which permits for flexibility and innovation throughout AI and HPC workloads.
OCP Open Rack Design
Constructed on the OCP Open Rack Broad design, “Helios” helps prospects and companions pace up deployment and gives a versatile answer for demanding AI workloads. This has enabled HPE to include distinctive applied sciences for its prospects, particularly a scale-up Ethernet change and software program made particularly for “Helios.” Developed in partnership with Broadcom, the change provides enhanced AI efficiency by means of the Extremely Accelerator Hyperlink over Ethernet (UALoE) normal.
Powering Herder Supercomputer
AMD can be powering Herder, a brand new supercomputer for the Excessive-Efficiency Computing Heart Stuttgart (HLRS) in Germany. It options AMD Intuition MI430X GPUs and next-generation AMD EPYC CPUs, delivering high efficiency and effectivity. Constructed on the HPE Cray Supercomputing GX5000 platform, Herder will present glorious efficiency and effectivity for HPC and AI workloads at scale.
Herder is designed to advertise scientific discovery and advance Sovereign AI initiatives throughout Europe. It should help complicated simulations and AI mannequin coaching inside a totally open, trusted infrastructure. By combining AMD’s main computing portfolio with HPE’s dependable system design, Herder might be a robust software for scientific analysis and industrial innovation for European researchers and companies.
