A deep dive into Nvidia's Vera CPU and the Olympus cores that power it

Nvidia has introduced its standalone CPU, Vera, directly challenging Intel and AMD's dominance in the CPU market. Designed for hyperscalers and cloud providers, Vera comes with 88 custom Armv9.2 cores, 176 threads, and support for up to 1.5 TB of LPDDR5X memory. It's a standalone platform independent of Nvidia's GPUs. The chip was unveiled in a whitepaper, shedding light on its inner workings, which reveal a unique architecture aimed at reducing pipeline and execution bottlenecks.

This design is intended to enhance performance in AI workloads and as a host for AI agents, which do not run on GPUs. The chip features a monolithic compute die, surrounded by various chiplets for I/O, memory, and connectivity. Vera supports both single- and dual-socket configurations, with the latter, dubbed the Vera CPU Superchip, featuring two CPUs connected via NVLink-C2C.

This configuration offers a total of 176 cores and 352 threads, with 16 SOCAMM2 LPDDR5x memory modules providing 2.4 TB/s of aggregate memory bandwidth. Nvidia's custom ARMv9.2-compatible core, Olympus, builds on existing Arm IP but features a custom neural branch predictor capable of exploring two branches simultaneously. This innovation aims to improve performance in branch-heavy workloads, such as Python scripts generated by AI code assistants.