Surprisingly Useful AI Article Enhancements
AMD has used the Open Compute Project (OCP) Global Summit in San Jose to showcase “Helios,” a rack-scale reference design for Ai infrastructure built on Meta’s new Open Rack Wide (ORW) specification.
The design, unveiled on October 15, represents AMD’s first public display of a system built to ORW standards, which were contributed to the OCP community by Meta.
Meta’s ORW Vision
Meta’s ORW specification defines an open, double-wide rack optimised for the power, cooling and serviceability needs of Ai-scale data centres. The design is intended to help hyperscalers and enterprises break away from proprietary infrastructure by creating a standardised and interoperable foundation for next-generation Ai workloads.
AMD’s Helios system brings that specification into practice. It is powered by the company’s Instinct MI450 GPUs, EPYC CPUs and Pensando networking technology.
Building Open Infrastructure
Forrest Norrod, Executive Vice President and General Manager of AMD’s Data Center Solutions Group, said Helios was a direct outcome of industry collaboration.
“Open collaboration is key to scaling Ai efficiently,” he said.
“With Helios, we’re turning open standards into real, deployable systems — combining AMD Instinct GPUs, EPYC CPUs and open fabrics to give the industry a flexible, high-performance platform built for the next generation of Ai workloads.”
The Helios design integrates multiple open compute standards, including OCP DC-MHS, UALink and Ultra Ethernet Consortium (UEC) architectures. These allow both open scale-up and scale-out fabrics to support high-throughput, low-latency connections across clusters.
Technical Features
The rack incorporates quick-disconnect liquid cooling to sustain thermal performance in large-scale deployments. Its double-wide layout is designed to improve serviceability, while standards-based Ethernet provides multi-path resiliency for data traffic.
AMD said the platform is purpose-built for frontier Ai and high-performance computing workloads, capable of delivering up to 1.4 exaFLOPS FP8 performance and 31 TB of HBM4 memory in a single rack.
Enabling Industry Adoption
As a reference system, Helios provides a template for OEMs, ODMs and hyperscalers to adopt, extend and customise Ai racks more rapidly. The company said this approach can reduce deployment times, improve interoperability and support efficient scaling of Ai and HPC environments.
By aligning with OCP standards, AMD aims to encourage wider adoption of open, standards-based infrastructure and provide organisations with more flexibility in how they design and run their data centres.
The debut of Helios underlines AMD’s intent to play a central role in building the infrastructure needed for large-scale Ai systems, while giving the broader industry a deployable design that bridges silicon, system and rack levels.
Last Updated on October 15, 2025 by Nick Ross
