Nvidia has confirmed Rigel, the custom Arm core that will power Rosa and drag its AI platform roadmap beyond Rubin.
The next CPU architecture after Olympus surfaced in an official Nvidia technical blog, where the outfit sketched its longer-term AI computing plans.
The current Vera CPU platform uses Nvidia’s custom Olympus Arm v9.1 architecture and scales to 88 CPU cores. Vera is due to pair with the upcoming Rubin GPU architecture as Nvidia’s next AI platform.
Rosa carries that plan onward with Rigel, a next-generation custom ARM v9.2 CPU core built to sit alongside the future Feynman GPU architecture.
Nvidia said Rigel is meant to deliver higher performance per CPU core while keeping roughly the same silicon footprint as Olympus. Rather than bloating the die, the company claims the gains come from architectural tweaks.
Those tweaks include better instruction delivery, a larger L2 cache and more efficient memory handling. The point is to improve compute efficiency for AI inference, training and other high-performance computing workloads.
Nvidia has not coughed up detailed Rosa specifications yet. Core counts, total cache capacity, memory subsystem design, manufacturing process and operating frequencies are still being kept in the corporate cupboard.
The company has not supplied performance estimates or benchmark comparisons between Olympus and Rigel either, which is handy if you prefer your roadmap with fewer awkward numbers.
The roadmap confirms Rosa processors will form the CPU side of Nvidia’s Feynman platform, which follows the Rubin generation now being developed. Rubin and Vera are the next big step in Nvidia’s AI infrastructure, while Feynman and Rosa push the plan further out.
According to Nvidia’s published roadmap, Feynman systems are expected to appear in 2028. The confirmation of Rigel gives an early look at Nvidia’s post-Rubin AI platform strategy, even if most of the juicy technical details remain missing.
