Correction: GB300 NVL72 delivers 1.6x performance on DeepSeekV3 pretraining at 512 GPU scale. (Ref: results 6.0-0022 and 6.0-0101)

→ View original post on X — @nvidia