Nvidia’s new Blackwell chips trained Meta’s large Llama 3.1 model in just 27 minutes. They used fewer GPUs and were over ...

Nvidia and its partners were the only entrants that submitted data about training that large model, and the data showed that ...