Skip to content
LandGadget
Computers & AINews

AMD Helion is an open rack blueprint for agentic AI, with up to 32,768 CPU cores

At the OCP Global Summit AMD introduced Helion, an open rack design for AI agents. The CPU-only version uses 6th Gen EPYC 'Venice' chips, and an all-in-one version adds Instinct GPUs.

LandGadget Staff · · 1 min read

Share
AMD Helion rack-scale system
Image: AMD
On this page

At the OCP Global Summit, AMD introduced Helion, an open blueprint for full server racks built for agentic AI. These are systems in which many AI agents plan, call tools and run code, and that work leans heavily on CPUs as well as GPUs.

Two configurations

  • CPU-only: sixth-generation EPYC "Venice" processors, scaling up to 32,768 cores per rack. It can be air- or liquid-cooled and is aimed at running large numbers of agents.
  • All-in-one: EPYC CPUs paired with Instinct MI350P accelerators in the same rack, so inference and agent execution can share the hardware.

Open by design

AMD says Helion is based on Open Compute Project specifications and that it will contribute the design back to OCP. Partner-built systems are planned from November 2026.

Why it matters

Training needed GPUs above all. Agentic workloads add a large amount of ordinary CPU work: orchestration, sandboxes and tool calls. AMD's move, together with Intel's announcements at the same summit, shows that the data-centre race is shifting from single chips to complete, standardised racks.

Source: AMD Newsroom — https://newsroom.amd.com/news/open-rack-blueprint-scale-agentic-ai/

Share

Keep reading