NVIDIA Launches Vera CPU for Agentic AI
NVIDIA Blog · rss · 2026-07-07
NVIDIA introduced "Vera," a next-generation CPU designed for agentic AI. The core claim is that for massive agent workloads, single-thread performance is more critical than simply piling on cores.
The rationale is that an agent's tool calling, code execution, data processing, and result verification all happen on the CPU, typically in serial pipelines. A slow single-thread CPU directly bottlenecks the agent loop, reducing GPU utilization and the overall ROI of the AI factory.
Key specs for Vera include:
- Custom Olympus cores, claiming a 50% IPC improvement over Grace
- Up to 1.2TB/s LPDDR5X bandwidth, with memory power consumption under 40W
- 3.4TB/s inter-core bandwidth, touted as 3x that of other data center CPUs
- Under full agentic workloads, it claims 1.8x sustained single-thread performance compared to x86
The article also cited benchmark results from Perplexity, Starburst, and Redpanda:
- Perplexity's real-world coding workflows are roughly 1.5x faster
- Concurrent sandbox startup is up to 1.9x faster
- Large-scale SQL analysis is 3x faster
- Real-time stream processing latency is up to 6x lower
Finally, NVIDIA mentioned a roadmap for its next-generation Rosa CPU and Rigel cores, continuing to advance CPU design for the agent era.
More from Infra
- Nearly half of Claude revenue now flows through cloud marketplace commitments, up from 11% — BenBajarin · 2026-10-26
- Anthropic ARR goes vertical after price cuts: tokens up 4x, $100B+ ARR back in play — GavinSBaker · 2026-10-26
- Alex Banks argues power caps GPU supply, real risk is demand cooling first — thealexbanks · 2026-10-26
- Analyst: Ant has the most fungible AI workload of any model provider today — BenBajarin · 2026-10-26
- Anthropic S-1: training infra edge cut per-task cost 40% and doubled revenue growth — BenBajarin · 2026-10-26
- Rewriting scaling law study: optimal synthetic data share rises from 10% to 32% with compute — heghbalz · 2026-10-26