AMD FSR antialiasing discussion

  • Thread starter Deleted member 90741
  • Start date
Someone with 6800XT and Godfall need to benchmark the game at various resolution and try to match the FPS at that various FSR preset.
I want to see what we can get without FSR quality wise vs with FSR for matching FPS.
I do understand that probably running FSR will have a performance penalty so running without FSR should be faster, but I'm curious on not only how FSR result compares with native but with its pre-FSR resolution.
Of course we can wait for the source code, which should be coming soon.
 
One potentially nice thing with AMD's solution besides the ability to support more games. Since it's not temporally based, it shouldn't introduce temporal artifacts. However, when combined with temporal rendering, it'll be interesting to see how it holds up.

The question is going to be, is it just good enough? Is it really good? Or is it like Quincunx and will introduce other annoyingly visible artifacts? (like Quincunx as in introducing very unpleasant artifacts, not in what it's doing :p)

Regards,
SB
 
interesting article. However, those people saying it cant compete with DLSS don't get that quality has zero to do with adoption

The closest example I can think of is FXAA and MLAA. A lot of developers included these “blurry AA” options in their games presumably because the AA benefits outweighed the loss in texture detail.

A similar calculation could be made for FSR given the significant performance benefits. Hopefully the availability of FSR will encourage developers to include more expensive/advanced rendering techniques in their games and raise the bar across the board.
 
The closest example I can think of is FXAA and MLAA. A lot of developers included these “blurry AA” options in their games presumably because the AA benefits outweighed the loss in texture detail.
Back then the choice was between no AA and these two so it was easy to justify. With FSR the choices are quite a bit more complex.
A simple example of recent days: will UE5 integrate FSR? Or do they consider their TSR enough/better at the same job?
 
FSR vs CAS would be interesting imo. It's not the same thing but the goal is the same, less pixel to render, more fps
CAS was applied on top of in-game spatial upscaling and FSR is AMD's spatial upscaling.
I guess they will continue using CAS on top of FSR, so the delta between the two will depend on how much AMD's spatial upscaling is better in comparison with in-game spatial upscaling (I am expecting just sharper geometry edges with better high res appearance).
 
Well, just like the RT, its something.

AMDs DXR implementation is functionally complete and as far as I’m aware it matches Nvidia for quality. The only difference is performance.

Upscaling us different in that there’s no reference implementation so it’s going to be very subjective.
 
Back then the choice was between no AA and these two so it was easy to justify. With FSR the choices are quite a bit more complex.
A simple example of recent days: will UE5 integrate FSR? Or do they consider their TSR enough/better at the same job?

This will get major adoption on systems like the aya and for people with lower end gaming laptops.

I also suspect just like Nvidia's solution this will get better as time goes on
 
If a spatial algorithm that was usable on all recent GPUs and was simple to integrate could provide almost native quality I would imagine we would have already had developers implement it. The more i think about it I will be shocked if it isn't noticeably worse than native.
 
If a spatial algorithm that was usable on all recent GPUs and was simple to integrate could provide almost native quality I would imagine we would have already had developers implement it. The more i think about it I will be shocked if it isn't noticeably worse than native.

If there is discernable degradation to image quality then the workaround is to provide a higher internal resolution so that we can feed more higher quality information to our filter. Fundamentally, all reconstruction techniques like upsampling are dependent on the amount of information we give them. In AMD's defense, there's valid reasons to not use temporal data since it's a lower quality source of information which introduces artifacts and requires a more implementation work ...

A spatial-only filter can still potentially beat the ML/temporal based filter in terms of quality and robustness if we supply it with better information ...
 
If there is discernable degradation to image quality then the workaround is to provide a higher internal resolution so that we can feed more higher quality information to our filter. Fundamentally, all reconstruction techniques like upsampling are dependent on the amount of information we give them. In AMD's defense, there's valid reasons to not use temporal data since it's a lower quality source of information which introduces artifacts and requires a more implementation work ...

A spatial-only filter can still potentially beat the ML/temporal based filter in terms of quality and robustness if we supply it with better information ...
Providing higher internal resolution just eats away at the performance benefit though. Can it beat what DLSS 2 is doing at comparable performance uplift over native?
 
A spatial-only filter can still potentially beat the ML/temporal based filter in terms of quality and robustness if we supply it with better information ...

For a technique that it is said to be “easy to integrate” and that doesn’t use temporal information there is not much room for being supplied with better information, whatever that means in practice.

Once you start lowering the rendering resolution there is only so much a spatial filter can do. I suspect the general assumption is that many games use TAA anyway and that it will mostly take care of temporal stability, which otherwise tends to be poor for superres spatial filters.
 
Providing higher internal resolution just eats away at the performance benefit though. Can it beat what DLSS 2 is doing at comparable performance uplift over native?

With a comparable performance uplift between them ? Possibly not because DLSS 2 has the advantage of having more information at hand even though some of it is not necessarily the best source. FSR trades this information away for more robustness so you never have artifacts like ghosting or post-process effects (DoF) shimmering and lower integration cost ...

Right now, I think it'd be an interesting proposal to have FSR covered on all quality metrics compared to what DLSS 2 offers even if it's higher quality setting does eat into the potential performance gains. Providing FSR with a higher internal resolution could still make for a very compelling option since it would be able to at least match DLSS 2 from a quality standpoint but also beat from a consistency/stability standpoint too!
 
With a comparable performance uplift between them ? Possibly not because DLSS 2 has the advantage of having more information at hand even though some of it is not necessarily the best source. FSR trades this information away for more robustness so you never have artifacts like ghosting or post-process effects (DoF) shimmering and lower integration cost ...

Without any temporal filtering the amount of flickering at lower res, magnified by the spatial filter, is often severe. With TAA enable to address these problems ghosting will re-enter into the equation. There’s no free lunch, as usual.

Right now, I think it'd be an interesting proposal to have FSR covered on all quality metrics compared to what DLSS 2 offers even if it's higher quality settings does eat into the potential performance gains. Providing FSR with a higher internal resolution could still make for a very compelling option since it would be able to at least match DLSS 2 from a quality standpoint but also beat from a consistency/stability standpoint too!
Never say never but I still have to see a temporally stable/consistent superres filter that is also spatial only, without the help of a temporal filter. Said that, I am ready to be surprised :)
 
Back
Top