AMD FSR antialiasing discussion

[troyan]

AMD suggests using FSR with lower texture mip bias and I would love to see how it affects temporal stability (would be a very bad idea to do something like this in a game with MSAA for example), neighborhood clipping in TAA is good at cutting off high frequency signal, but any texture noise that survive should be quite apparent and can be futher exaggerated by RCAS.

Marvels Avengers shows the same temporal instability like Terminator: Imgsli
It is really bad.
Game doesnt support a simple resolution scale. But even driver upsampling from 1080p -> 2160p with nVidia doesnt introduce the negative lod artefacts: https://imgsli.com/NjE1Nzc

 

[OlegSH]

Marvels Avengers shows the same temporal instability like Terminator:

Thanks for the screenshots. I don't have the game, so I can't check by myself.
That shimmering on traffic signs is quite visible even on static screenshots.
Yeah, if texture bias was all good and safe, everybody would just slap -3 of it in native resolution, but nobody does this since there are obvious trade offs.

 

[DieH@rd]

https://twitter.com/i/web/status/1416146545788006409

FRS + Temporal AA on PC and PS5. Since the game has cartoney art style, the loss of texture quality [when the frame is rendered at lower res before FSR is applied] will be less noticeable.

 

[pharma]

They also have some video shots of Marvel Avengers ...
AMD FSR and NVIDIA DLSS comparison shows similar quality and performance at 4K resolution in Necromunda Hired Gun - VideoCardz.com

 

[troyan]

The prolog of Avengers is just bad for FSR. You can even disable TAA with FSR - a shot between 4K/FSR UQ/DLSS Q: Imgsli

Sharpness is the real difference between native and FSR upscaling. Reconstruction of the image is miles behind native.

 

[DegustatoR]

Seem like they are using FSR and TAAU simultaneously in Arcadegeddon if only on PS5:

https://twitter.com/i/web/status/1416171637645271040

I wonder for what purpose though. Seems like doing just TAAU with some CAS on top of it would result in a better IQ with similar performance gains.

 

[pjbliverpool]

Additionally, Marvel Avengers has added FSR support and also supported DLSS when it came out. Same comparison method: FSR Ultra Quality vs DLSS Quality.

https://imgsli.com/NjE0ODU
https://imgsli.com/NjE0ODc
https://imgsli.com/NjE0ODg
https://imgsli.com/NjE0ODk
https://imgsli.com/NjE0OTI
https://imgsli.com/NjE0OTM
https://imgsli.com/NjE0OTQ

Performance is about the same for both techniques in this game. Quality..., well, judge by yourself.

That's a very easy win for DLSS. Just look at the cables on the bridge/sea pick. And DLSS is faster too, albeit marginally. FSR is still doing a great job though, I'd be more than satisfied with that image quality if I needed the extra performance. I'd probably leave it off if I already had 60+ fps though.

 

[OlegSH]

https://drive.google.com/drive/folders/1l4iMdM0xrF1-mWpR5E8LprP7w6jjpNQQ?usp=sharing
Comparison screenshots in Necromunda Hired Gun.
"DLSS Quality sharpness 20" screens are DLSS Quality + additional sharpness to make sharpness levels the same for FSR and DLSS.
By default textures with FSR are sharper than both native and DLSS, hence the adjustment. That sharpenning filter is not free though, it adds up some frame time on top of the default sharpening in DLSS.
Difference in small details handling is huge, it's very visible that FSR is an upscaling in this game. DLSS looks better than native, but lacks sharpness that really helps in highlighting all the small details in this game.

 

[pharma]

 

[Malo]

What's with the lights on the tower at 0:50-1:00? In both methods the lights are all over the place. Is that chaos present in non-upscaling? It's terrible.

 

[Remij]

What's with the lights on the tower at 0:50-1:00? In both methods the lights are all over the place. Is that chaos present in non-upscaling? It's terrible.

Seems like an effect meant to help make the player feel like they are moving fast while flying... despite moving slow as molasses lol

 

[gamervivek]

DLSS in Avengers seems to have some issues with post-processing, in below comparison you can see big differences on the building.

https://imgsli.com/NjE1Nzg

And the bloom on DLSS image is muted, even if you ignore the Iron man boosters, the lights on top of the building look like as if the bloom is missing.

There was similar difference in death stranding, though not as striking as here. So wonder if it's simply inherent to DLSS.

https://www.nvidia.com/en-us/geforce/comparisons/death-stranding-nvidia-dlss-2-0-comparison-004/

 

[orangpelupa]

DLSS in Avengers seems to have some issues with post-processing, in below comparison you can see big differences on the building.

https://imgsli.com/NjE1Nzg

And the bloom on DLSS image is muted, even if you ignore the Iron man boosters, the lights on top of the building look like as if the bloom is missing.

There was similar difference in death stranding, though not as striking as here. So wonder if it's simply inherent to DLSS.

https://www.nvidia.com/en-us/geforce/comparisons/death-stranding-nvidia-dlss-2-0-comparison-004/

The trees also looks sharper on DLSS. Maybe it inherently skips / removes blurry stuff?

 

[Dictator]

DLSS in Avengers seems to have some issues with post-processing, in below comparison you can see big differences on the building.

https://imgsli.com/NjE1Nzg

And the bloom on DLSS image is muted, even if you ignore the Iron man boosters, the lights on top of the building look like as if the bloom is missing.

There was similar difference in death stranding, though not as striking as here. So wonder if it's simply inherent to DLSS.

https://www.nvidia.com/en-us/geforce/comparisons/death-stranding-nvidia-dlss-2-0-comparison-004/

It depends on where and When they do their Bloom probably, before or after dlss. OG Pixel size changes the size of bloom

 

[nikobellic]

There was similar difference in death stranding, though not as striking as here. So wonder if it's simply inherent to DLSS.

Its like TAAU in UE4 that has similar issues like these, it seems like temporal upsampling does correctly with some post process effects like bloom or DOF, and its not a simple issue to fix. Maybe UE devs were able to fix it on TSR (UE5) but on TAAU thats present in UE4, this type of issue is still present.

 

[troyan]

Someone did a movement comparision in Necromunder:

https://www.reddit.com/r/nvidia/comments/om790k

The same movement artefacts are visible in this game, too. Unlike normal upscaling FSR is introducing more artefacts and it seems that game developers dont care to provide better alternatives like bilinear upscaling combined with the RCAS filter.

 

[TalEth]

https://drive.google.com/drive/folders/1l4iMdM0xrF1-mWpR5E8LprP7w6jjpNQQ?usp=sharing
Comparison screenshots in Necromunda Hired Gun.
"DLSS Quality sharpness 20" screens are DLSS Quality + additional sharpness to make sharpness levels the same for FSR and DLSS.

Lol if you need to add more sharpening to DLSS (with the added perf penalty ), that is already a big win for FSR.

 

[DavidGraham]

In conclusion, NVIDIA’s DLSS implementation in Marvel’s Avengers is truly amazing. Not only does it look better than AMD’s FSR, but it also runs faster. Additionally, DLSS reduces aliasing to such a degree that is even better than native resolutions.

https://www.dsogaming.com/pc-perfor...-native-4k-versus-nvidia-dlss-versus-amd-fsr/

 

[OlegSH]

Lol if you need to add more sharpening to DLSS

I would not call "a win" something that causes posterization, i.e. the painterly look (and other artifacts too).

Unfortunately, many people have no notion of how the texture detallisation is different to sharpening. They mistake sharper image for better details, so that additional sharpening is just to highlight texture details that already present in the image with DLSS for such people.

 

[DegustatoR]

Adding sharpening to DLSS in comparison to FSR is fine simply because FSR has RCAS built into itself. I wish that every DLSS2 game would provide an option of controlling DLSS sharpness though.