Nvidia announced DLSS 2.0 - got my hopes back up regarding this technology, results looks promising from their commercial demonstration at least.
Yeah I like the idea and if they can simplify the requirements which from the sounds of it they have this has potential to see further adaption with a bonus to additional customization options being exposed as well and both improving performance and image quality also which still had some room for improving even if from what I've been hearing and also seeing in comparison videos NVIDIA has already improved things quite a ways since the tech was introduced although the one downside is that the earlier titles would have to be re-trained against the newest algorithm making updates less likely although not impossible. (Aside from Final Fantasy XV I suppose which basically had the PC version support come to a abrupt halt but it happens even if that was a bit faster than intended.) Additional benchmarks and tests on Control and later on Mech Warrior 5 should make for a good initial showcase too when these updates launch, I might be on AMD myself but I still like the idea and technology and improving it further would eventually bring it very close to native high resolution display and it's advantages but at a less costly performance level and that's not bad at all. Temporal enhancements and improving in motion visual quality would be a big step towards this as I'm seeing it nobody wants the effect to "break" or well worsen whenever anything on the screen is in motion after all including many effects and other elements as well. (Much as I know about it at least which well it could be better but I'm learning a little here and there, somewhat. ) EDIT: Guessing the performance hit alone also means full offline scaling even if less visually impressive is out so no means at least anytime soon for a override in the driver directly but that might have been overly optimistic on my part considering other scaling means clearly show that even accelerated via say CUDA it's still a lot of work for processing. This is still good though and getting more game engines involved could also see the effect seeing wider adaption and usage having the brunt of the implementation covered for already. Although with every game needing some AI learning and training that part might be the tougher for having this supported at least in the amount of time needed and getting full quality from the result. (Starting involvement early could help with that though, what do I know anyway nice update here! )
Looking forward to Control update featuring this updated DLSS, if any game should showcase an improvement this is clearly this one. Arguably their former approach of having to train their model 'in screen space' requiring training on full game picture and specific resolutions sounds a lot like a 'brute force' approach. I'm glad they moved away from it as it wasn't making any sense - DLSS should have been trained only on edges not full pictures: the work intended by the algorithm is to smooth out edges - being temporal or static - if you only train the model to solved edges, you don't need to be game specific nor resolution specific which appears to now be the case with DLSS 2.0. We've been stuck with TAA for far too long - this technology has a massive performance cost, creates ghosting and blurriness - it needs to go. Pretty confident AMD will come up with an alternative to DLSS. I wouldn't be surprised if DLSS end up being a Nvidia Control panel option in the coming months. Nvidia has been actively working with Unreal Engine for years now, with specific version of unreal engien available on github - so the number of game benefiting any of its technology has been growing steadily for a while now.
