That seems surprising? I would expect datacenters GPUs are already optimized for that, since it's one of the main difference between how datacenters and gaming GPUs are evaluated.

It should have been trivial to find a source if the assumption was true, which makes me doubt even harder.