NVIDIA's Blackwell Ultra architecture (GB300) is slated for 2026 production, promising 1.5 petaFLOPs of FP4 AI compute per chip β a 50% increase over the B100. With 288GB of HBM3e memory and NVLink 5 interconnects, the GB300 NVL72 server rack is expected to become the new standard for frontier AI training runs at the $50-100 billion compute scale.

Comments on "NVIDIA Blackwell Ultra GPUs (GB300)"
Create a free account or sign in to join the discussion.
Sign in to join the conversation