As you can see, there are rarely told about artifacts that are obvious coming from that frame-by-frame masking technology, especially flickery change of global lights, shadows, textures, LOD objects. Watching him playing it makes it nauseating. And even if he had twice the specs he have now, these crawling changes to the picture would be faster, but they won’t be any more consistent.
NVIDIA packed a lot of side-products to promote their hardware, and from all of them DLSS5 is their worst hill they will to die on. While some like Broadcast enabled streamers to clear their sound channel or improve visuals, sampling tech and hyper resolution tricks made demanding games a bit more performant (if ever), this fancyfier only gobbles resources while providing a picture that’s not only counter to what artists wanted to show, but just plain incomprehensible when seen in motion.
And it’s not like we can wait for DLSS6 to fix that. The graphics pipeline behind this technology is built to fail at everything but a still screenshot. But it would still trend as long as there would be CEOs and directors who don’t play their own games and can’t tell the difference, or there wouldn’t be enough bad press that is more probable.
Jesus, this looks pretty bad. And the best parts are nowhere nearly good enough to justify the average, let alone the worst parts.
Horrid.
Ideally the effect isn’t applied universally across the scene, it’s supposed to be configured by devs so different elements of a scene get different treatment. Retrofitting it to a game is going to produce the worst possible results compared to a more narrow targeted use where its intentionally applied by artists during the dev process.
I don’t see a clear and obvious reason why they won’t model e.g. faces initially the way they want it to be and instead involve a process that applies a rather unpredictable mask over the scene.
I can see engine developers seeing gamers now having all these additional raytracing and tensor cores collecting dust and jumping at the opportunity to put them to work, possibly lowering the load on other hardware parts. E.g. instead of populating the landscape with debris, grass, trees procedurally via CUDA-like cores draw them over bare land, add detales that weren’t there on objects where consistency is not needed.
But for now it lack any thought behind it. The tool is there but it has not found a purpose. And it needs to find one fast, because there is a short window before it gets written off as a chip-melting waste of frames. A more universally applicable and useful raytracing technology is still not a standard, there is like one Doom game that doesn’t have baked in lighting at all, and DLSS5 stays on even lower platform. We need to see at least one good example of it.
