3 ms·
i hope this might offer the feature for AMD cards, or something unified so that this feature could be utilized in games without having specific hardware.
by chii 5y ago
i hope this might offer the feature for AMD cards, or something unified so that this feature could be utilized in games without having specific hardware.
- ChuckNorris89 5y agoHow? Nvidia's lawyers will be all over that shit. Not AMD and no respectable OSS devs will touch it, nor will GitHub or any other reputable platform host it. It's legally radioactive. Nintendo's HDL and source code also leaked from a hack a couple of years ago but given how litigious Nintendo is, nothing came out of it despite everyone expecting some emulators to pop op.
- userbinator 5y agoIt's called "don't ask, don't tell" and "parallel construction".
- ChrisRR 5y agoI'm sure nVidia's lawyers will respect "don't ask"
- userbinator 5y agoThat's when the parallel construction comes in. Obviously anyone copying the code verbatim or close to it deserves to get the lawyers coming after them, but plausible deniability should be strongly kept in mind for people doing RE+reimplement work anyway.
- xxs 5y agoif anything - it makes it worse. The code is untouchable for any serious company or even open source. If you need references see what WinNT/XP code leaks did to Wine project.
- dncornholio 5y agoAMD has FidelityFX and it's open source :) Maybe they can borrow some of the DLSS techniques
- SketchySeaBeast 5y agoFidelityFX leverages an open source spatial upscaling algorithm - it doesn't integrate any sort of machine learning. I don't know if you could bridge the gap without radically changing what it's doing, especially in terms of now needing to provide training data.
- kllrnohj 5y agoFSR doesn't use things like motion vectors which are more likely to be the quality difference between it and dlss then machine learning is. Temporal AA & upscaling is well known to produce fantastic results without any machine learning involvement. Hence why dlss 2.0 doesn't need per game training anymore. The "machine learning" part of it seems to be mostly killed off unless they somehow managed to train a fully generic model that can handle all art designs and game engine settings (eg, motion blur on vs. off). But it seems more likely it's a fancy generic take on temporal upscaling, possible influenced by ML but not dominated by it. Which I guess is something we'll find out with this leak. How much AI is actually in dlss 2.0 and how much is "just" offloaded temporal super sampling (such as epic's TSR)