AMD FSR antialiasing discussion

  • Thread starter Deleted member 90741
  • Start date
I'd take FSR over checkerboarding any day. Thats why i think FSR is such a good addition to the PC gaming space, its the checkerboarding of the consoles but better.
 
That's great news and a great feather in FSR's cap IMO. The one area that REV falls short in vs the console versions (aside from the DRM issue which is being resolved) is in it's checkerboard upscaling technique which I understand is pretty poor vs the console version. I would hope that FSR can do a much better job and at least equalise things in that regard, if not give the PC version the advantage. It'd be nice to see Alex do an update to the current REV face off once FSR is in place and the DRM issue has been patched out.
No, problem with REV village is that that the TAA is broken in the PC version and much less effective compared to console version. The checkoarding technique is actually pretty good.
 
No, problem with REV village is that that the TAA is broken in the PC version and much less effective compared to console version. The checkoarding technique is actually pretty good.

If that's the case then FSR may not help at all. Shame the game doesn't support DLSS which would replace the TAA altogether.
 
Looks like the number of games supported by FSR increased a little bit.
Almost every game supported by Proton in Linux can use FSR now. Good news for future Steam Deck users.

https://github.com/ValveSoftware/wine/pull/116


Here it is running on Forza Horizon 4:



In the meanwhile someone has been working on implementing FSR through a shader injector:
https://github.com/Blinue/Magpie


I think it's only a matter of time until a method to apply FSR 1.0 into virtually anything will come to fruition, be it through the GPU's driver (of any IHV) or through an unofficial patch.
 
I think it's only a matter of time until a method to apply FSR 1.0 into virtually anything will come to fruition, be it through the GPU's driver (of any IHV) or through an unofficial patch.

FSR seems to be a perfect fit for something like Reshade. It's basically a post process filter. The question is how does it get the app to request a framebuffer at the upscaled resolution.
 
FSR seems to be a perfect fit for something like Reshade. It's basically a post process filter. The question is how does it get the app to request a framebuffer at the upscaled resolution.
But you really don't want something like FSR integrated as a post process filter as it's imperative that game post-processes like bloom, DOF, CA etc, and HUD of course are applied after FSR.
 
Good news for future Steam Deck users.

Are Steam Deck users going to be playing games at 4K, with or without FSR? As long as FSR doesn't involve a sub-sampling optimization then I see no reason to be optimistic about using it in cases where your input buffer has inadequate spatial sampling (low resolutions, non-anti-aliased images.) It feels like FSR right now is cashing cheques that were written by some of DLSS 2.0's remarkable results at lower resolutions.
 
Are Steam Deck users going to be playing games at 4K, with or without FSR? As long as FSR doesn't involve a sub-sampling optimization then I see no reason to be optimistic about using it in cases where your input buffer has inadequate spatial sampling (low resolutions, non-anti-aliased images.) It feels like FSR right now is cashing cheques that were written by some of DLSS 2.0's remarkable results at lower resolutions.

Nah, plenty of people with handheld consoles are pretty happy with using FSR.
Just because it doesn't look as good at lower resolutions doesn't mean it's useless..

And the good thing about FSR is that they can use it on Intel and AMD based handheld PCs.
There are zero handhelds with RTX GPUs.
 
Are Steam Deck users going to be playing games at 4K, with or without FSR? As long as FSR doesn't involve a sub-sampling optimization then I see no reason to be optimistic about using it in cases where your input buffer has inadequate spatial sampling (low resolutions, non-anti-aliased images.) It feels like FSR right now is cashing cheques that were written by some of DLSS 2.0's remarkable results at lower resolutions.
We're talking about a 4 Watt mobile solution with shared memory here. It's pretty clear, that this is not the stuff that will run CP2077 in 4K ultra-max-epic-RT. As long as it does not introduce too many artifacts of it's own, I welcome the option to choose whether or not it's a net-win for me on a game-per-game basis wrt quality vs. playability.
 
Watched a few Resident Evil videos on youtube and FSR is just bad. Do not see any "advanced" upscaling.
With Avengers the driver upscaling from 1440p to 2160p is around 21% faster than FSR Quality (not performance!) in 4K on my 3090. "Image quality" is slighty better with FSR but it has the same movement artefacts. Do not see any reason to chose it over a simple bilinear upscaling. It would be a shame if developer will only support FSR going forward. 17% less performance for basically no real improvement isnt worth it.

/edit: Captured with Geforce Experience:
FSR: https://streamable.com/4ab13k
1440p: https://streamable.com/atx9cu
 
Last edited:
Computerbase agrees with this analysis:

Ultimately, however, the problem of blurring in the game is so great that even FSR should only be activated on "Ultra Quality" in emergencies and then only in Ultra HD. The combination produces a very useful result, but it is no longer possible to speak of comparable image quality. With FSR she suffers visibly - always. A comparable image quality cannot be achieved with FSR in Resident Evil Village. Here it may be more sensible to reduce individual graphic options first.

Here FSR simply stumbles on the game's own TAA, which unlike Nvidia's competing technology DLSS, has to be used with FSR. And since the game's TAA reacts to a reduced rendering resolution with massive blurring, FSR is automatically blurred as well. And that already applies to the highest FSR setting "Ultra Quality" in Ultra HD.
https://www.computerbase.de/2021-07/resident-evil-village-amd-fsr-test/2/
 
seems DLSS uses way too much imagination on the speakers

View attachment 5746

EDIT:

the blotchy pattern on the side of the speaker also changed a bit in DLSS
Something is not quite right in that video. Here is the original video that pic came from:
Start around 43 seconds. Released on 7-18

Here is another video showing the same scene although the character isn't in the exact same spot:
Start around 2:09

As you can see the speaker grill is fine. Released on 7/20. And, the second video is a bit further away from the 1st video.
At a guess this appears to be a game engine problem that may have been fixed with the released of some starter pack on 7/19.

If it's to be assumed/guessed that the 1st video is correct representation of the grill it is done at native using TAA option in the menu. Which would tell me that TAA is the actual problem. FSR only makes it worst do to the final render. Which would explain to me why, in this example, DLSS disables TAA (as it's grayed out in most games). While using their own implementation of it through DLSS 2.1

The real question is: Why is native resolution, while using TAA, not showing the grill as intended?

Edit:
I can only hope that we don't start a trend of comparing FSR with crap, low quality TAA while DLSS disables TAA in the menu using their own.
 
Last edited:
https://videocardz.com/newz/amd-fid...-fp32-fallback-tested-native-fp16-is-7-faster


What do you propose to compare FSR with? It doesn't really work without some kind of AA which is TAA in most games.
Although I find your question disingenuous as I made it quite clear that TAA in Avengers was garbage. It did make me ponder on it further.
  • These results are using a 3080ti and 3090. Not Radeon Cards. So, is the result of grill mesh not rendering properly do to a driver issue? That is still an unknown right now.

And to answer your question FSR should also use it's own post processing solution. That would override in game AA just like DLSS.
This is using a 6900xt btw



Necromunda Hired Gun: FSR enabled which disables TAA. Which suggest that FSR is using their own post processing when using a Radeon 6900xt





Necromunda Hired Gun when DLSS is enabled. You can clearly see that AA is grayed out.





Avengers: FSR is only using the garbage TAA from Avengers.

As you can see above this is exampled from Necromunda Hired Gun were that's exactly the case. And, the reason why FSR and DLSS nearly look identical (i'm only reference the Radeon Card not the nvidia cards used as I noticed anomalies). And why FSR looks like garbage when comparing Avengers (garbage aa implementation).

Now, I don't pixel peek (zooming still images at 400-500x zoom). I take the image AS IS based on game play. In other words when formulating an opinion on FSR we have to make the distinction of what GPU is actually being used. As their maybe difference, albeit subtle, between competing GPUs when FSR is used.

Therefore, to better explain my response:
I can only hope that we start to see a trend of comparing FSR/DLSS with their own implementation of PP which overrides in game AA as exampled in Necromunda Hired Gun. Because making any IQ comparison where FSR is reliant on in game, low grade TAA vs DLSS using it's own makes an invalid comparison IMO. Which is found in Avengers.
 
Last edited:
Back
Top