The FSR Thread (Announcements & Discussion)

a 4K 55 inch w/120hz freesync oled is fine here :drool:
it's wall mounted about 4 feet away

i think a 65 inch in 8k would be good

Used to think that would be too big... now not so sure... at 4 feet must feel a little like an Imax lol.
 
FSR 2 looking pretty good. Strangely he says that DLSS had a slight edge and in the video it seemed like DLSS had far more obvious trails in motion. But they seem pretty close. Just gotta hope devs add it to more games now...
 
Looks like dlss still has some ghosting but the fsr2 had ever so slightly less detail in motion but its always when they blow it up 300% which no one would notice.

I call it a draw between the 2 which is pretty amazing considering they dont use any ai with FSR.
 
how long will DLSS last now ? :hmm:

since it only works on RTX cards not even all NV cards

Plus FSR will work on consoles. I expect within a few months we'll be seeing EVERY major release supporting FSR. I really think this is a sea-change moment, and we'll be seeing quite a bump in image quality that will be very noticeable on current-gen consoles.

Don't count out DLSS. They're working on their 3.0 version, and there's likely to be some really great improvements. At this point though, I think we're going to be hitting diminishing-returns for upscaling image quality. But there's other areas that DLSS can push that will be meaningful.
  • Ease of implementation. nvidia needs to move away from game-specific training sets & models to a general model. Game-specific training is a GIGANTIC amount of work for devs (rumor is that nvidia almost always has to do it themselves). The existence of competitive technologies like FSR 2.0 that don't require any training at all... Honestly I think DLSS 3.0 will be DOA if it still requires game-specific training. I've already seen people successfully applying newer games DLSS models to older games and seeing some improvements there, so this is entirely doable if nvidia wants to.

  • Performance. If nvidia can improve image quality for lower-resolution rendering, that would be a major competitive advantage.

  • Compatibility with all modern nvidia cards. This means building fall-backs for older hardware, and could theoretically even allow them to run on competitors' cards if they chose to. None of this sounds like stuff nvidia will want to do. But as their competitors are using "ubiquity" (including consoles) as their competitive advantage, nvidia will need to find some way to stay competitive.

  • AI upscaling has the potential to grow into doing lots of graphical things that FSR can never do. Theoretically, AI upscaling could change the graphics style of a game: stylize your game into a cartoon? make it look like an old-timey film? Add a level of photo-realism? Researchers are currently doing tons of crazy stuff with AI-based images, and all of those are theoretically possible for DLSS to grow into.
 
Plus FSR will work on consoles. I expect within a few months we'll be seeing EVERY major release supporting FSR. I really think this is a sea-change moment, and we'll be seeing quite a bump in image quality that will be very noticeable on current-gen consoles.

Don't count out DLSS. They're working on their 3.0 version, and there's likely to be some really great improvements. At this point though, I think we're going to be hitting diminishing-returns for upscaling image quality. But there's other areas that DLSS can push that will be meaningful.
  • Ease of implementation. nvidia needs to move away from game-specific training sets & models to a general model. Game-specific training is a GIGANTIC amount of work for devs (rumor is that nvidia almost always has to do it themselves). The existence of competitive technologies like FSR 2.0 that don't require any training at all... Honestly I think DLSS 3.0 will be DOA if it still requires game-specific training. I've already seen people successfully applying newer games DLSS models to older games and seeing some improvements there, so this is entirely doable if nvidia wants to.

  • Performance. If nvidia can improve image quality for lower-resolution rendering, that would be a major competitive advantage.

  • Compatibility with all modern nvidia cards. This means building fall-backs for older hardware, and could theoretically even allow them to run on competitors' cards if they chose to. None of this sounds like stuff nvidia will want to do. But as their competitors are using "ubiquity" (including consoles) as their competitive advantage, nvidia will need to find some way to stay competitive.

  • AI upscaling has the potential to grow into doing lots of graphical things that FSR can never do. Theoretically, AI upscaling could change the graphics style of a game: stylize your game into a cartoon? make it look like an old-timey film? Add a level of photo-realism? Researchers are currently doing tons of crazy stuff with AI-based images, and all of those are theoretically possible for DLSS to grow into.

Didn't nvidia ditch game specific training like two years ago with the release of dlss 2.X?
 
Didn't nvidia ditch game specific training like two years ago with the release of dlss 2.X?

Oh, looks like you're correct. I didn't realize they had moved away from per-game training.

Because temporal artifacts occur in most art styles and environments in broadly the same way, the neural network that powers DLSS 2.0 does not need to be retrained when being used in different games. Despite this, Nvidia does frequently ship new minor revisions of DLSS 2.0 with new titles, so this could suggest some minor training optimizations may be performed as games are released, although Nvidia does not provide changelogs for these minor revisions to confirm this.

https://en.wikipedia.org/wiki/Deep_learning_super_sampling

Well that's good.
 
Very pleased to see Fsr 2.0 -- 4k quality mode offer comparable image quality in some respects to Dlss and offer a flexible sharpening slider.
 
Annoying, Farming Simulator 22 doesn't have the update with FSR 2.0 yet on the XBox app :(
 
Back
Top