Nvidia announces “Rubin Ultra” and “Feynman” AI chips for 2027 and 2028
1 min read
Summary
Nvidia has revealed details about its forthcoming AI-accelerating GPUs at its GTC 2025 conference, including the Vera Rubin chip, which is named after the famous astronomer, and features tens of terabytes of memory as well as a custom-designed CPU called ‘Vera’.
Due for release in second half of 2026, Rubin Ultra will deliver 3.6 exaflops of FP4 inference compute, while the latter, set for release in the second half of 2027, will deliver 15 exaflops of FP4 inference and 5 exaflops of FP8 training performance, offering around four times more power than the Rubin NVL144 configuration.
The company said the GPUs would help to deliver major performance improvements for AI training and inference, with the chips’ architecture specially designed with AI-oriented instructions to make it easier for developers to write efficient AI algorithms.
