NVIDIA's Vera CPU Targets Faster, More Efficient AI Agents in Data Centers
If NVIDIA's own and partner-reported figures hold up, agent workloads could run on fewer CPUs per unit of power, though none of the numbers have been independently benchmarked.
NVIDIA says its Vera CPU is built for AI agents, which constantly hand work between GPU reasoning and CPU execution, and that it can run more agents per unit of power in AI data centers. In a 100-megawatt data center comparison, NVIDIA reports one Vera CPU delivers 1.6x more agents per watt than the fastest x86 CPU and 2x faster agent speed than the most energy-efficient x86 CPU. Early partners report gains in their own tests: Perplexity says agentic coding ran 1.5x faster, DeepInfra reports up to 2.2x speed over the best x86 processor, and Daytona says it brought 1,000 agent sandboxes online in 10.8 seconds versus 88.4 seconds on x86. Perplexity plans to deploy Vera in an upcoming production system. Figures come from NVIDIA's own comparisons and partner-reported tests rather than independent benchmarks.
Why it matters
The claims matter most to operators of AI data centers, where agents repeatedly pass work between GPUs and CPUs, so the CPU's speed and power draw affect how much agent work a facility can run. The strongest evidence so far is the comparison with x86 processors, but it comes from NVIDIA itself, and the partner results come from tests the partners ran on their own workloads. Perplexity's planned production deployment is the first sign of whether these gains show up outside controlled comparisons.
Matters or hype?
Does this story matter, or is it hype? Vote before you see what everyone else said.
Sources · 1
- NVIDIA Blog ↗ Primary source