Nvidia Unveils Vera Rubin GPU Roadmap at GTC 2025
Nvidia reveals its next-generation GPU lineup at GTC 2025, headlined by the powerful Vera Rubin chip slated for late 2026. The new roadmap also includes Blackwell Ultra for later this year and the distant Feynman architecture arriving in 2028.
Nvidia CEO Jensen Huang reveals an extensive new GPU roadmap at the GTC 2025 conference in San Jose. The flagship announcement is the Vera Rubin architecture, which pairs a new custom Nvidia CPU called Vera with a powerful dual-GPU setup. Slated for release in the second half of 2026, Vera Rubin promises to significantly outperform the current Grace Blackwell generation.
Vera Rubin delivers up to 50 petaflops of performance for AI inference tasks, which is more than double the 20 petaflops offered by the current Blackwell chips. The accompanying Vera CPU also operates at roughly twice the speed of the processor found in the Grace Blackwell system. Looking slightly further ahead, Nvidia plans to launch Rubin Ultra in the second half of 2027, packing four GPUs into a single package to achieve up to 100 petaflops of performance.
For the near term, Nvidia prepares to release Blackwell Ultra in the second half of 2025, maintaining 20 petaflops of AI performance while increasing memory from 192GB to 288GB. On the far horizon, the company teases the Feynman architecture, named after the famous physicist, which is scheduled to arrive in 2028 and will also utilize a Vera CPU. This multi-generational roadmap shows Nvidia's aggressive strategy to maintain its dominance in the AI hardware market.