NVIDIA details Rubin GPU roadmap, promises 3.3x training speedup over Blackwell

NVIDIA used GTC to unveil the Rubin GPU family, the architectural successor to Blackwell. The headline number: 3.3x faster pre-training on a like-for-like cluster.

  • Rubin (R100) GPUs ship in H2 2026
  • HBM4 memory, 8 stacks per package
  • NVLink Switch 7th gen, 1.8TB/s GPU-to-GPU

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top