NVIDIA Vera CPU Aims To Boost Agent Speed And Per-Watt Capacity

AI agents now do far more than draft responses: they search for information, query databases, write and verify code, and trigger actions. Much of this work depends on the CPU, with agents moving between GPU-based reasoning and CPU-side execution throughout a task.

As organisations scale up deployments, data centres need CPUs that can drive these steps quickly while supporting large numbers of concurrent agents within strict power budgets. The objective is to increase useful work per watt without lengthening queue times.

NVIDIA positions its Vera CPU as delivering both high single-agent performance and strong agent density per watt in one processor. The company argues this lets infrastructure providers standardise on a single CPU across varied agent workloads, instead of mixing "fast" and "dense" x86 parts to balance speed and capacity.

In a 100-megawatt data centre comparison, NVIDIA reports that a single Vera outperforms both the fastest x86 CPU and the most energy-efficient x86 option. According to the company, Vera delivers industry-leading agent speed while enabling 1.6x more agents per watt versus the fastest x86 CPU, and provides 2x faster agent speed with leading energy efficiency compared with the most energy-efficient x86 CPU.

Early adopters say they are seeing similar results in production-style testing:
- Perplexity: Ran agentic coding tasks 1.5x faster with Vera than on traditional CPUs, and plans to deploy Vera in an upcoming production system.
- DeepInfra: Achieved up to 2.2x the speed of the best x86 processor on agentic orchestration for its production agent workload, with Vera the fastest of four architectures tested.
- ClickHouse: Its standard binary, with no Vera-specific tuning, scaled near-linearly across 176 cores and topped the ClickBench hardware leaderboard.
- DuckLabs: Recorded about 1.5x the median TPC-H SF1000 performance of an Intel Xeon 6 baseline.
- Daytona: Brought 1,000 agent sandboxes online in 10.8 seconds using Vera versus 88.4 seconds with x86, and sustained 12.7 completed jobs per second at 2,000 concurrent sandboxes.
- Modal: Saw up to 2x sandbox throughput per core compared with comparable x86 hardware, with gains increasing under heavy load.

www.nvidia.com/en-gb/