The development
What happened
NVIDIA says Vera uses 88 Olympus CPU cores, with each core delivering 50% higher instructions per cycle than Grace. The company specifies up to 1.2 TB/s of LPDDR5X memory bandwidth at under 40 watts of memory power and 3.4 TB/s of core-to-core bandwidth. NVIDIA claims 1.8 times the sustained per-core performance of x86 in loaded agentic workloads; Perplexity reports about 1.5 times faster completion of a coding workflow. NVIDIA also previews its next-generation Rosa CPU with Rigel, an Arm v9.2 core with improved instruction delivery, a larger L2 cache, and more efficient memory handling.
Why it matters
If independent testing supports the claims, Vera’s per-core performance could shorten CPU-bound tool calls and data-processing steps inside agent workflows.
What the reporting establishes
- Vera combines 88 Olympus cores with a monolithic compute die.
- NVIDIA specifies up to 1.2 TB/s of LPDDR5X memory bandwidth and 3.4 TB/s of core-to-core bandwidth.
- The 1.8-times sustained per-core comparison is an NVIDIA claim against x86 systems.
- Perplexity reports roughly 1.5-times faster completion of its cited coding workflow.
- Rigel is an Arm v9.2 successor core planned for the future Rosa CPU.
What remains unclear
- The performance figures come from NVIDIA and named partners; the supplied reporting does not provide a complete independent benchmark methodology.
- NVIDIA does not provide system pricing or a release schedule for Rosa and Rigel in the supplied material.
Source reporting
- Tom's HardwareIndependent reportingNvidia touts Vera CPU's single-threaded performance as its agentic AI advantage, reveals next-gen 'Rigel' Arm CPU cores — frames chip as a 'max single-threaded CPU at scale,' not a parallel monster
- NVIDIA BlogPrimary sourceAI Innovators Adopt NVIDIA Vera — Why Max Single-Threaded CPU at Scale Matters