AI infrastructure teams are no longer optimizing only GPU throughput; agent systems spend real time in CPU-bound tool calls, code execution, retrieval, and verification loops. NVIDIA used a July 7 blog post to position Vera as a data-center CPU designed for those agentic workloads, saying its Olympus cores, memory bandwidth, and monolithic die are meant to keep per-core performance high under load. The company says Perplexity tested Vera on a coding workflow involving repository cloning and test execution, completing the job about 1.5x faster than x86 and starting concurrent sandboxes up to 1.9x faster. NVIDIA also previewed its next Rosa CPU with Rigel cores. For LDS readers, the practical signal is that inference economics are broadening beyond accelerator choice into CPU, memory, and software co-design for autonomous-agent stacks.
CoreWeave Deploys Vera Rubin, Integrates Training and Inference