Why are they only testing inference vs training?

Not many companies are going to want to deploy their own public-facing chatbot service. But almost everyone in this space is going to want to train their models, which is where the performance boost comes in.

This is the big iron competition that’s going to get really interesting.

If amd overtakes nvidia in server sales…

Though I think demand is so high both are going it be sold out at capacity for the foreseeable future.