Nvidia Executive Rejects Claims of Predicting Agentic AI During Vera CPU Development

Nvidia is pushing back against criticism that its Vera CPU was specifically designed to capitalize on the current wave of agentic AI workloads. Ian Finder, head of Nvidia’s data center CPU business, told Tom's Hardware that it would be "absurd" to claim the company predicted the agentic moment four years ago when development on the chip began.

Finder argued that the design of the Vera CPU, which is driven by AMD, is based on design intuition and a co-design approach with Nvidia accelerators rather than a specific prediction of future AI trends. He stated that the company's strategy involves designing CPUs to keep accelerators fed and happy, a process that began with the Grace data center CPU. This approach utilizes an "opinionated" microarchitecture known as the Olympus core and an opinionated memory system via the Scalable Coherency Fabric.

According to Finder, Nvidia's "clean sheet of paper" approach allows the company to design hardware without the legacy constraints or dollar-per-core economics that affect competitors like Intel and AMD. This method prioritizes high single-thread performance, concurrency through IPC rather than frequency, and high memory bandwidth. While Finder noted that the company saw the beginnings of tool calling four years ago, he maintained that the team did not explicitly design the chip around the specific agentic workloads seen today.

Nvidia's strategy aims to capture market share in a CPU space that is accelerating due to AI agents. While benchmarking shows the Vera CPU does not excel at every workload and faces competition from AMD's EPYC Venice, Finder suggested that Vera's superior memory bandwidth will become increasingly important as agentic systems evolve to better utilize available compute.

FI Journal
Privacy Overview

This website uses cookies so that we can provide you with the best user experience possible. Cookie information is stored in your browser and performs functions such as recognising you when you return to our website and helping our team to understand which sections of the website you find most interesting and useful.