NVIDIA unveiled Vera, a new CPU class designed specifically for agentic AI workloads, delivering 1.8x the sustained per-

NVIDIA introduced Vera, a new category of CPU built for the agentic AI era. The company positions max single-threaded performance at scale as critical for AI agents, which operate in continuous loops where each step depends on previous results. While traditional data center CPUs have evolved toward higher core counts and lower costs, they sacrifice per-core speed. Vera reverses this trend, optimizing for individual core performance across all 88 cores without bottlenecks.

At Vera's core is Olympus, NVIDIA's custom CPU core delivering 50 percent higher instructions per cycle than NVIDIA Grace. The CPU pairs these faster cores with up to 1.2TB/s of LPDDR5X memory bandwidth using less than 40 watts of memory power. Critically, its monolithic compute die provides 3.4TB/s of core-to-core bandwidth, triple that of any other data center CPU, allowing all cores to access full memory performance without slowdowns.

In loaded CPU workloads representing agentic execution, Vera delivers 1.8x the sustained per-core performance of x86. Perplexity tested Vera on real agentic work including repository cloning and test suite execution in sandboxes, finding Vera completed the job approximately 1.5x faster than x86 and started concurrent sandboxes up to 1.9x faster. Perplexity plans to deploy Vera in its upcoming production system.

Beyond tool execution, Vera accelerates data workloads common to agents. Partners measured 3x faster large-scale SQL analytics with Starburst and up to 6x lower latency on real-time streaming with Redpanda compared to leading x86 server CPUs. Vera serves as a unified architecture across the AI factory, powering the NVIDIA Vera Rubin GPU system and the BlueField-4 STX storage processor on the same platform and toolchain.

NVIDIA's next-generation Rosa CPU featuring the Rigel core will continue this roadmap. Rigel delivers higher per-core performance than Olympus while maintaining the same silicon footprint, with improvements in instruction delivery, larger L2 cache and more efficient memory handling. The company frames Vera as foundational for an era where billions of agents will depend on CPUs for execution, verification and data retrieval, maximizing GPU utilization for revenue-generating work.