Nvidia Unveils Next-Generation Rosa CPU for AI Agent Workloads

According to Nvidia, the company recently unveiled technical details of its next-generation Rosa data center CPU, designed specifically for emerging AI agent workloads. Rosa will be paired with the next-generation Feynman GPU and will adopt the new Rigel CPU core, Nvidia's latest custom Arm microarchitecture following Olympus, both based on Arm v9.2 instruction set.

Unlike Vera, which uses the Olympus core, Rigel focuses on enhancing single-core performance within the same silicon area rather than increasing core count. Key improvements include more efficient instruction dispatch, larger L2 cache, and more efficient memory access mechanisms. The expanded secondary cache and improved data access capabilities are intended to reduce memory access frequency and boost execution efficiency for AI inference tasks involving frequent branching and data scheduling.