DLSS is built on NGX - DirectML is essentially an alternative framework to NGX. Nvidia had to build NGX because DirectML didn't exist. The rest of the stuff you wrote I'm aware of but doesn't have anything to do with NGX vs DirectML or the argument of whether Nvidia could have done DLSS in a way that AMD could benefit. I don't think anything is necessary until it is - I mean at some point a caveman got tired of being a caveman when everyone else was fine with it for millions of years. We had a technology, machine learning, that was being used for all kinds of image processing - so Nvidia decided to pivot that technology into games. They built an SDK, which does more than just DLSS and said "let's let developers see what they can do with machine learning in video games". It could have lead to better AI, DLSS could have been gods gift to gaming, it could lead to things I can't even think about - it could lead to nothing. Either way I'm a fan of innovation - it drives the industry forward. NGX drove Microsoft to create/finish DirectML - it drove AMD to pursue DLSS alternatives on DirectML. So regardless to whether it was necessary at the start or not it's pushing developers out of caves and potentially starting something great. This is something that Nvidia has done for a while now regardless to whether gamers like them or not. AMD's GPUOpen alternatives are all in response to Gameworks. AMD wasn't talking about Freesync or using vblank to sync frames to the GPU prior to Gsync. AMD wasn't investing in ray collision hardware until Nvidia pushed RTX. Obviously it goes both ways, with AMD pushing Nvidia on certain things (Async Compute) - I wish Nvidia would be open with their tech and libraries but regardless it's still innovative and drives the industry forward.