we have nVidia who clings selfishly to their proprietary blobs, and I can’t help but wonder how great it could be if they opened that up and let the community in.

Nvidia is doing that because they don’t want people to deploy gaming GPUs in datacenters, and they can currently enforce that through their driver license. That license is what enables them to force most enterprise users to buy expensive A100/H100 datacenter GPUs and rack in really fat margins when a couple of RTX 4090 cards would actually be enough to do the job with good cost efficiency. The control that Nvidia has with that license is not something they’re ready to give up and that’s why they keep giving the middle finger to the FOSS community.

(before anyone mentions vast.ai as a counter-example, those RTX 4090 compute sellers are indeed breaking Nvidia’s EULA)