MLPerf Training v6.1 is out.
@CoreWeave is the first cloud provider to deliver multi-rack
@nvidia Vera Rubin results in MLPerf Training v6.1.
Our DeepSeek-V3 671B submission reached target quality in 17.65 minutes on 256 GPUs across four racks, and our Llama 3.1 8B submission was the fastest at 72 GPUs.
Same production infrastructure our customers get - great work team đ! $CRWV