This is exactly why I'm worried. I took an ML class last semester and having access to CUDA on my 1080Ti (explicitly purchased for gaming) was a boon for my pro

This is exactly why I'm worried. I took an ML class last semester and having access to CUDA on my 1080Ti (explicitly purchased for gaming) was a boon for my project group, since we could actually tune models efficiently. This might be good for the short-term, but it sets a precedent I'm really not happy with. I know they already do some of this with Quadro, but my understanding there is that the value add is in its reliability and certification, not ability to operate on workloads that have been artificially crippled on the gaming cards.