CoreWeave $CRWV posted the first multi-rack Nvidia $NVDA Vera Rubin results from a cloud provider in MLPerf Training v6.1.
It trained DeepSeek-V3 in 17.65 minutes on 256 Rubin GPUs, 2.7x faster than its own GB200 result.
It finished Llama 3.1 8B from $META in 7.16 minutes on 72 Rubin GPUs, 1.9x faster than its GB200 run.
On GB200, it also ran reinforcement learning on Qwen3.5 397B from Alibaba $BABA in 22.78 minutes on 256 GPUs.
顯示更多