AMD Helion Press Release
AMD Helion rack-scale configurations support diverse AI agents that plan, execute, and manage thousands of concurrent tasks. Configure AI Your Way.
Not All Agents are the Same
For personal agents, the priority is maximum agents per watt; for enterprise agents, it’s the maximum number of high-performing agents; and for technical compute agents, it’s maximum performance per agent.
Not All Agentic Rack Needs are the Same
For hyperscale deployments, CPU-only disaggregated racks execute agents; for enterprise deployments, all-in-one racks combine CPU and GPU in a single system for agent execution at scale.
Configure it Your Way
AMD Helion is an open, rackscale reference design purpose-built for agentic AI and powered by AMD EPYC™ 9006 Series Server CPUs. Choose from a CPU-only hyperscale configuration or an enterprise configuration with a mix of CPU and GPU blades, both built on OCP standards. Select the configuration and blade options that best match your agent needs.
CPU-only agentic rack
Hyperscale Configuration
A CPU-only rack designed for hyperscalers, neoclouds, and sovereign AI operators deploying agents at large scale and optimizing for number of agents per watt. Mix and match CPU blades within the rack to support your diverse agents per watt.
Personal Agents
Dense Servers
256c @ 400W
Enterprise Agents
Balanced Servers
256c @ 600W
Technical Agents
Performance Servers
96c High Frequency
CPU + GPU agentic rack
Enterprise Configuration
Intelligently route your agents’ inference requests to frontier models running in the cloud or open-source models running inside the same rack. The enterprise configuration uses similar infrastructure to the CPU-only hyperscale configuration, with a mix of CPU and GPU blades to match your agent needs.
Agentic CPU Blade
Offers high-density and high-performance options for agents running at high volume and technical compute agents that need fast, low-latency.
Dense Servers
AMD EPYC™ 9006
256c @ 400W
Balanced Servers
AMD EPYC™ 9006
256c @ 600W
Performance Servers
AMD EPYC™ 9006
96c @ 5.0 GHz
GPU Inference Blade
Adds GPU inference to the rack to support demanding agentic workflows.
Performance Servers
AMD Instinct™ MI350P
General-Purpose Blade
Supports the database, storage, and technical computing workloads that agents rely on to complete their work.
Database and Storage Servers
AMD EPYC™ 9006 SP7 and 9006 SP8 Server CPUs
Technical Compute
AMD EPYC™ 9006X SP7 Server CPUs
Specifications
Built for High-Volume Agentic Workloads
Up To
32,768
"Zen 6" x86 CPU cores
Up To
16
AMD Instinct MI350P GPUs
Up To
102
Tbps of networking bandwidth
21” ORv3 44OU DLC and air-cooled rack, Native OCP Compliant
Fully programmable to support diverse, mixed agent fleets
AMD Helion Partners
Adopted by leading OEMs and ODMs
Frequently Asked Questions
The needs of hyperscalers and enterprises are fundamentally different. Hyperscalers deploy agents at very large scale, optimize for number of agents per watt, and already operate separate accelerated infrastructure — a CPU-only rack is the efficient answer. Enterprises typically want a single rack that delivers both agentic compute and acceleration without standing up two separate racks.
Yes — that is the design intent. Mix and match CPU blades to match your agent mix, and in the Enterprise configuration add GPU blades alongside them.
AMD Instinct™ MI350P, eight GPUs in a 5U server. Two GPU servers can be added to a rack for sixteen GPUs total — two nodes of eight. The two nodes operate independently; there is no scale-up GPU fabric between them, because this configuration is designed for enterprise agentic workloads and acceleration rather than frontier-scale training.
Orv3 for the rack and busbar. DC-MHS for the host processor module. DC-SCM for management and security. HPRv2 for power shelves. OCP NIC 3.0 for networking. Caliptra for the silicon root of trust. That is compliance at every layer, not a rack frame that happens to be OCP-shaped.
No — they serve different needs. AMD Helios™ is built for frontier-scale AI training and large-scale inference with a unified scale-up GPU domain. AMD Helion Enterprise configuration is for organizations that want agentic infrastructure with inference in a single rack.
Resources
