Semiconductor company AMD said its sixth-generation EPYC 9006 “Venice” server processors are in production, with computer manufacturers preparing to launch systems and leading cloud providers expected to begin deployments later this year.
The processors support agentic AI, or AI systems that can carry out a series of tasks to complete a request. An AI agent may search company records, use a business application, run code, and combine the results into an answer. Each step puts different demands on the servers running it.
AMD’s approach is to offer different processor configurations for those tasks. Some need faster processing to deliver a quick response, while others need enough capacity to handle many requests at once. The company’s processor portfolio spans smaller deployments with eight cores to flagship chips with 256 cores. Cores are the processing units within a chip that carry out computing work.
The shift matters because planning AI infrastructure now involves more than providing computing power to train models or generate answers. AI agents also rely on databases, web services, and other business software as they work through an assignment.
A server suited to one step may be less suited to another. AMD said its four processor families share a common software foundation, allowing customers to choose chips for different jobs while using the same underlying software environment.
In a new white paper, AMD evaluated the Venice processors across enterprise applications, cloud services, AI, scientific computing, and general-purpose tasks.
AMD reported that its Epyc 9996 processor delivered 1.2 times the performance per core and 2.24 times the overall platform performance of an Nvidia Vera-based system in SPECrate 2026 Integer testing, a benchmark for comparing computing performance.
The company also reported performance gains ranging from 2.4 to 3.7 times across enterprise and cloud software tests. These included MongoDB and Redis, which help applications store and retrieve data; NGINX, which handles web traffic; and transaction processing.
Against Intel’s Xeon 6980P processor, AMD reported performance advantages of 1.8 to 3.13 times in tests involving molecular simulations, materials modeling, and weather forecasting.
AMD also estimated that its processors could deliver 3.4 times the throughput of a Vera-based platform in a modeled 100-kilowatt server rack. That figure is a projection of how much work the rack could process, rather than a result from an operating customer installation.
The findings come from AMD’s testing and modeling. For businesses building AI agent systems, the company’s pitch centers on matching processors to the different tasks within a workflow as computing requirements change.