AMD executives question Nvidia, Arm's "smart CPU": no one could predict the wave four years ago

Abstract:

According to Tom’s Hardware report,

Two AMD executives recently questioned Nvidia and Arm’s description of the new processors as “designed for intelligent AI.”

Rangarajan said that AMD believes that the sixth-generation EPYC 9006 series is a server CPU suitable for the era of agents, but the agent workflow is not a single task, but covers a variety of loads such as model invocation, tool execution, data retrieval, code running, and task orchestration. Therefore,

processors need to cover different computing needs rather than relying on a single model to solve all problems

.

AMD Zen chief architect Mike Clark also said that the company would not claim to have foreseen the wave of intelligent computing in the early stages of Zen 6 development. However, AMD has previously judged that CPUs will still play an important role in the AI ​​software ecosystem. With the rapid development of agent applications, the existing general-purpose computing power of the Zen 6 architecture can also be used for related workloads.

These remarks point to Nvidia and Arm’s recent positioning of data center CPUs. When Nvidia released the Vera CPU in March this year, it called it the world's first processor "designed for the era of intelligent AI and reinforcement learning." This product uses 88 self-developed Olympus cores, which are mainly responsible for intelligent agent tool invocation, task orchestration and data processing.

Arm also positions the AGI CPU as a data center processor for agent AI, saying that its operating frequency, memory and I/O architecture are all designed around large-scale parallel agent tasks. AMD maintains that intelligent AI requires a complete product portfolio composed of different types of CPUs.

The differences between the two parties mainly focus on the two technical routes and product narratives of "dedicated design" and "universal architecture adaptation".