Nvidia DLSS 1 and 2 antialiasing discussion *spawn*

[DegustatoR]

I've pointed out other reviewers claiming different.

You've posted DLSS 1.0 vs CAS review which is how AMD proposed to compare them back at 5700 / CAS launch.

Even in the video you posted about DLSS they compared it to ps4 pro checkerboard but not radeon fidelity cas + upscaling.

Because Fidelity CAS upscaling is just that - upscaling. It doesn't reconstruct anything and thus has a natively worse resulting image quality. There are no comparisons of CAS upscaling with DLSS or TAA SS or CBR because of that. It's worse than any reconstruction.

 

[Leoneazzurro5]

There is NO way (and it's about the one-millionth time someone has to say it) an upscaled image looks better than native resolution. It may look better than a native resolution with tons of TAA applied on it, which blurries textures a lot, that is clear. There are tons of comparatives out there other than the DF one showing that DLSS 2.0 provides an upscaled image that has a quality "in the middle" between the native resolution and the lower resolution used as a base.

 

[DegustatoR]

There is NO way (and it's about the one-millionth time someone has to say it) an upscaled image looks better than native resolution. It may look better than a native resolution with tons of TAA applied on it, which blurries textures a lot, that is clear. There are tons of comparatives out there other than the DF one showing that DLSS 2.0 provides an upscaled image that has a quality "in the middle" between the native resolution and the lower resolution used as a base.

DLSS is antialiasing as well as reconstruction. Comparing it to TAA is valid.

 

[tsa1]

All these comparisons are very subjective. Just look at the debates around fonts. Many designers prefer subpixel anti-aliasing and heavy hinting that makes the picture "smoother" and less jaggy while a substantial fraction of users (about a third, myself included) simply can't read with such 'improvements' (as demonstrated when Chrome cut out GDI rendering) as it turns the text into a blurry mess that hurts the eyes. Some can tolerate it, some don't, same goes for the visual artifacts / a range fringe cases that each method of image denoising/upscaling creates.

 

[pjbliverpool]

There's no comparison between DLSS2 and CAS. At least not in Death Stranding:

https://www.overclock3d.net/reviews...tranding_-_nvidia_4k_performance_trump_card/3

 

[DavidGraham]

DLSS also introduces artifacts and removes some effects something that Cas doesn't do.

Old material, Hardware Unboxed visited the subject again with DLSS 2 and praised it significantly stating it holds a significant advantage over even native resolution. CAS doesn't even stand a chance.

 

[dorf]

There is NO way (and it's about the one-millionth time someone has to say it) an upscaled image looks better than native resolution. It may look better than a native resolution with tons of TAA applied on it, which blurries textures a lot, that is clear.

Rendering at 99% resolution + utilizing one previous frame cannot produce better results than rendering only the current frame at 100%? I'd argue the 1.98x the sample count of the first case compared to the second can produce better results even if half of the samples are less than ideal.

This might have come off as pedantic but just trying to make a point.

 

[Deleted member 2197]

Because Fidelity CAS upscaling is just that - upscaling. It doesn't reconstruct anything and thus has a natively worse resulting image quality. There are no comparisons of CAS upscaling with DLSS or TAA SS or CBR because of that. It's worse than any reconstruction.

Spoiler

Death Stranding And DLSS 2.0 Gives A Serious Boost All Around (wccftech.com)

 

[Leoneazzurro5]

Rendering at 99% resolution + utilizing one previous frame cannot produce better results than rendering only the current frame at 100%? I'd argue the 1.98x the sample count of the first case compared to the second can produce better results even if half of the samples are less than ideal.

This might have come off as pedantic but just trying to make a point.

What part of "upscaled" are you not getting? You are NOT using the 99% of the resolution + utilizing a previous frame, you are using 1/2.25 the sample count even in the "quality" preset at 4K, plus ML "magic". DLSS may give a quite small IQ hit for a decise boost in performance, and I agree to this. Repeating as a mantra the Nvidia marketing slogan "DLSS is better than native" does not make it true .

 

[Lurkmass]

I don't know why there's so much emphasis behind DLSS 2.0 or why AMD should develop a similar alternative when there less than 10 games that support it ? Most of the games that support DLSS are of older implementations and never got updated again ...

 

[DegustatoR]

I don't know why there's so much emphasis behind DLSS 2.0 or why AMD should develop a similar alternative when there less than 10 games that support it ?

https://www.rockpapershotgun.com/2020/10/20/confirmed-ray-tracing-and-dlss-games-2020/

 

[dorf]

What part od "upscaled" are you not getting? You are NOT using the 99% of the resolution + utilizing a previous frame, you are using 1/2.25 the sample count even in the "quality" preset at 4K, plus ML "magic".

Upscaling from 99% render resolution to full resolution is what I meant in my example. This is still upscaling.

Also, my point wasn't to argue DLSS exactly, just questioning your claim about upscaled image never being able to be better than native res when we have the added factor of differing image processing / temporal accumlation techniques for both cases.

 

[Jay]

What part of "upscaled" are you not getting? You are NOT using the 99% of the resolution + utilizing a previous frame, you are using 1/2.25 the sample count even in the "quality" preset at 4K, plus ML "magic". DLSS may give a quite small IQ hit for a decise boost in performance, and I agree to this. Repeating as a mantra the Nvidia marketing slogan "DLSS is better than native" does not make it true .

So are you saying that native at lower settings is always better than DLSS2 at higher ones?

As the main reason to use DLSS2 is to gain performance to use for higher fps and/or settings.

 

[Kaotik]

https://www.rockpapershotgun.com/2020/10/20/confirmed-ray-tracing-and-dlss-games-2020/

Is that list gonna hold? None of the promised RT/DLSS lists NVIDIA outed themselves did.

So are you saying that native at lower settings is always better than DLSS2 at higher ones?

At least native is free of DLSS2's artifacts, let alone app specific issues. I'm especially dumbfounded by the Death Stranding praise, with DLSS it has all the usual sharpening artifacts including halos, as well as moiré issues, broken particle effects and even those weird motion vector tails. And that's supposed to be the best and better than other options in the game?

 

[CarstenS]

People tend to like crispy images with punchy colors. Just show someone an unprocessed RAW file from a DSLR in comparison with the JPG the camera produces at the same time. Arguably, the RAW file has much more useful information in it and is closer to reproducing the real image, while the JPG is tuned to popular taste.

 

[Leoneazzurro5]

Upscaling from 99% render resolution to full resolution is what I meant in my example. This is still upscaling.

Also, my point wasn't to argue DLSS exactly, just questioning your claim about upscaled image never being able to be better than native res when we have the added factor of differing image processing / temporal accumlation techniques for both cases.

Native resolution is what the image is meant to be. Every upscaling, even at 99% of the resolution, is still upscaling, so it is a downsampling of said image and may lead to artifacts. So no, no upscaling can be "better than native". It may look as good as native in many cases, I already agreed to that. But it is still not the original and the claims that is "better than original" are laughable.

 

[Leoneazzurro5]

So are you saying that native at lower settings is always better than DLSS2 at higher ones?

As the main reason to use DLSS2 is to gain performance to use for higher fps and/or settings.

uh? I am saying that 4K native IQ without IQ-reducing postprocessing is better than a 4K DLLS-processed image IQ based on a lower resolution downsample like 1440p or 1080p, at the same settings regarding lighting, shadows and textures. Never said DLLS is not useful to increase FPS (like all other upscaling techniques).

 

[DegustatoR]

Is that list gonna hold?

Unless some games will patch their DLSS support out, there's 20 games with DLSS available right now. Which I think is more than "less than 10"?

 

[Jay]

uh? I am saying that 4K native IQ without IQ-reducing postprocessing is better than a 4K DLLS-processed image IQ based on a lower resolution downsample like 144op or 1080p. Never said DLLS is not useful to increase FPS (like all other upscaling techniques).

I'm just wondering why you would do that anyway.
You run DLSS2 for a reason, I've not seen anyone say they would run it if they was getting everything they required at native resolution. Not yet anyway, although can be situations where may look better.

It's also still early days even if it doesn't feel like it.

 

[Lurkmass]

Unless some games will patch their DLSS support out, there's 20 games with DLSS available right now. Which I think is more than "less than 10"?

The majority still don't use DLSS 2.0 so my point still stands regardless ...

One could make a bold claim that DLSS 2.0 doesn't really exist right now and they wouldn't be all that far off ...