Nvidia's Vera CPU Shines at Hot Chips 2026 with Agentic AI Performance

Nvidia's Vera CPU, designed specifically for agentic AI workloads, was showcased at Hot Chips 2026. The company highlighted its unique features, including a custom Nvidia core, spatial multithreading, and an LPDDR5X memory subsystem.

According to Nvidia, Vera is optimized for high single-core throughput, with large cores featuring a BPU, neural branch predictor, and 10-wide decode. Spatial multithreading allows threads to share resources within the core, reducing the impact of neighboring threads on performance.

In benchmarks against AMD's EPYC 9655P, Vera demonstrated significant improvements in agentic workloads such as headless browser usage and code compilation. Nvidia claims that Vera can compile the Linux kernel 22% faster with a native AArch64 target and 14% faster when cross-compiling for x86.

Nvidia also emphasized the importance of power efficiency, using LPDDR5X to reduce power consumption while maintaining high bandwidth. The company demonstrated that a fully loaded memory system with Vera consumes between 30W and 40W, compared to traditional RDIMMs which can consume over 100W.