← Latest articles

AMD Develops Verano Processor to Anchor Next-Generation AI Server Racks

AMD EPYC 9754 processors and memory in a dual-socket server system in a Geekerwan hardware video.
Photo credits
Photo: 极客湾Geekerwan · CC BY 3.0 · Cropped within the original photograph to 16:9, resized, and converted to WebP; no padding or synthetic background.

AMD is developing a new processor called Verano to serve as the host brain for its upcoming rack-scale artificial intelligence systems. The hardware is built specifically to supply data to AMD graphics accelerators, positioning the combined setup to compete directly with Nvidia’s Vera Rubin superchips.

In an interview with reporters in Las Vegas, Madhu Rangarajan, corporate vice president of AMD’s compute and enterprise AI products business, explained that Verano is engineered to function inside tightly integrated, rack-sized hardware setups. "Verano is really designed for an integrated CPU+GPU rack scale solution," he said. "From that perspective, I wouldn't say the CPU is a competitor to Vera. In a head note, I'd say the Verano+GPU solution is a competitor to Vera+Rubin."

Platform Architecture and Memory Trade-Offs

Nvidia’s competing platform joins an 88-core central processor and memory to a pair of graphics units on a single large board. Verano takes a similar approach by acting as a dedicated host processor that relies on notebook-style memory modules rather than standard server memory sticks.

This memory choice allows the system to consume 30 to 40 percent less power and fit into a more compact space for liquid cooling, though it uses a soldered connection layout that makes the processor harder to service. Unlike standard notebooks, the memory modules themselves are not soldered down and use the same modules found in Nvidia's superchips.

Internal Design and Performance Links

To keep the graphics accelerators running at full speed without data bottlenecks, AMD redesigned the processor's communication links. The chip uses custom connection channels that achieve speeds of 112 gigabits per second per lane.

Rangarajan noted that notebook-style memory can be sensitive to signal loss when passing through standard connectors, which led AMD to solder the processor directly to the board. "LPDDR tends to be quite finicky in terms of speeds," he said. "Every time you go through a connector, you tend to lose a pin or two, and you already have one connector with the SoCam itself. So to minimize any of those losses in speed, we went BGA on the CPU."

Under the outer cover, Verano shares a visual resemblance to AMD's Venice server processors and is offered in two configurations depending on customer workloads. Buyers can choose a version with 72 high-frequency cores reaching up to 5.3 gigahertz, or a denser option featuring 192 cores operating at up to 4.1 gigahertz.

While AMD has not yet disclosed the exact ratio of processors to graphics chips it plans to use, Verano is scheduled to begin shipping next year alongside the company's next-generation accelerator hardware.

Summary

AMD's Verano processor represents a structural shift in how the company builds server hardware for artificial intelligence, moving away from general-purpose computing toward tightly integrated rack systems that directly challenge Nvidia's dominant superchip architecture.