Nvidia DLSS 1 and 2 antialiasing discussion *spawn*

Are you sure you didn't misread the labels? The placement of the labels between shots makes it harder to determine whether the label is referring to the shot above or below it. For every pair, the top shot seems to be with DLSS on and the bottom one DLSS off (native). For the first pair:
DLSS: https://www.hardwaretimes.com/wp-co...Control-Screenshot-2020.03.25-19.28.04.46.png
Native: https://www.hardwaretimes.com/wp-co...Control-Screenshot-2020.03.25-19.28.28.30.png
Yeah I inspected the 1080p shots on their webpage and matched the names to the 4k downloads to be sure.
Are we seeing the same thing? first pictures are DLSS on, and those are the better ones, the softer ones are DLSS Off/native. Far away text is very slightly distorted with DLSS on of course, but the overall sharpness of the image speaks for itself.
It's subjective I guess but IMO there's very little difference in overall sharpness between the shots but there's much more distortion around edges on objects/characters with higher contrast/larger distance to the background object.

I'm sure it looks great in motion and the point of it is to come really close to the original image at a much higher performance in which without having access to it myself, it's seemingly doing just that. I just don't like statements like "it's better than native!" you purposely quote, it's tacky.
 
I'm sure it looks great in motion and the point of it is to come really close to the original image at a much higher performance in which without having access to it myself, it's seemingly doing just that. I just don't like statements like "it's better than native!" you purposely quote, it's tacky.
OK see here for what I mean:

DLSS ON: https://www.hardwaretimes.com/wp-co...Control-Screenshot-2020.03.25-19.35.41.72.png
DLSS Off: https://www.hardwaretimes.com/wp-co...Control-Screenshot-2020.03.25-19.36.02.56.png

With DLSS on, the patterns on the radio, the poster, the text on the coffee machine all appear sharper and more defined than native, that's what I mean by DLSS being better than native+TAA, that's a straight one in my books.

Same thing here, see the text on the elevator sign to the right and the portrait in the hall.
DLSS ON: https://www.hardwaretimes.com/wp-co...Control-Screenshot-2020.03.25-19.34.52.56.png
DLSS Off: https://www.hardwaretimes.com/wp-co...ntrol-Screenshot-2020.03.25-19.35.10.00-1.png
 
Again, are you looking at the right labels? I have seen the originals of the ones you mentioned, the DLSS picture is definitely sharper and more defined than the native one.
They didn't bother labeling the full size images, but since they're not exactly the same they can easily be recognized regardless of which looks better or worse.
Here's examples of what I described, both blown up and direct crops, in spoilers because of the size
Ringing artifacts present on DLSS (left), missing without (right):
upload_2020-3-25_20-25-46.png
upload_2020-3-25_20-25-30.png

Strange dark row of dots (there's more of them one hinge to the left which I cut out, also that cupboard edge looks awfully "rough")
upload_2020-3-25_20-27-54.png
upload_2020-3-25_20-29-40.png
Strange overemphasized "raster pattern" + messed up "rules"
upload_2020-3-25_20-31-25.png

upload_2020-3-25_20-31-38.png
 

Attachments

  • upload_2020-3-25_20-27-48.png
    upload_2020-3-25_20-27-48.png
    37.6 KB · Views: 8
Like before, you can set custom DLSS resolutions in Control in the renderer.ini file. They work up to around 93% resolution scale after which the reconstruction will break down into a mess.

neat - a shame it does not work at native like it did in 1.9.

Just tested it out - looks a lot better than 1.9, much more stable in motion. Transparencies also look a lot better.
 
DLSS 2X is different where supposedly it's using native resolution and superenhancing AA based on deep learning. It's only mentioned in the initial Nvidia PDF and never again. DLSS 1.0 and DLSS 2.0 are both using lower than native render resolutions.
 
Strange dark row of dots
These are not strange dots, these are the indirect RT shadows, they are often quite noisy, and the DLSS helps make them less noisy actually, and cleans up their noisy patterns.

Strange overemphasized "raster pattern" + messed up "rules"
While true, you can't help but notice that the text in the DLSS image is sharper and more visible.

Overall these are very minor blemishes, they are completely overshadowed by the large fps boost and the sharper overall image.
 
But sharpness is easy and cheap if you don't mind artifacts ... and since I can count the staircases on the textures we can obviously accept some more artifacts for the native image too.
 
and since I can count the staircases on the textures we can obviously accept some more artifacts for the native image too.
Yeah of course, we are now at a point where the user has to choose between two different rendering methods each with different set of strengths and weaknesses, with the caveat that one offers very large increase in performance over the other.
 
Like @Malo pointed out, I was talking about DLSS X2, not DLSS 2.0. DLSS X2 is the mode they advertised when they first introduced DLSS, where you start with your native resolution and DLSS tries to match up to the supposed ideal image (64x res/ SSAA or something I think it was?).
DLSS X2 is still M.I.A.

These are not strange dots, these are the indirect RT shadows, they are often quite noisy, and the DLSS helps make them less noisy actually, and cleans up their noisy patterns.
If they're indirect noisy RT shadows which DLSS helps cleaning up, why are they completely missing from the image without DLSS and present on the one with DLSS? Shouldn't it more likely be reversed then?
While true, you can't help but notice that the text in the DLSS image is sharper and more visible.

Overall these are very minor blemishes, they are completely overshadowed by the large fps boost and the sharper overall image.
Sharpness isn't by itself necessarily desired option, DLSS seems to go way over in many places which leads to strange things.
 
Last edited:
If they're indirect noisy RT shadows which DLSS helps cleaning up, why are they completely missing from the image without DLSS and present on the one with DLSS? Shouldn't it more likely be reversed then?
This goes back to the days of DLSS 1.9, DLSS enhanced the pattern of indirect shadowing (with more coverage), it was working better in tandem with DLSS than native res.
Sharpness isn't by itself necessarily desired option, DLSS seems to go way over in many places which leads to strange things.
Many in the tech press and many gamers seemed satisfied when AMD's sharpness solution was all the rage to the point of announcing it was better than DLSS 1, despite not being able to achieve native sharpness. What we have with DLSS 2 is even better than that IMO, native sharpness is achieved despite few blemishes with the added benefit of a large fps boost.
 
This goes back to the days of DLSS 1.9, DLSS enhanced the pattern of indirect shadowing (with more coverage), it was working better in tandem with DLSS than native res.

Many in the tech press and many gamers seemed satisfied when AMD's sharpness solution was all the rage to the point of announcing it was better than DLSS 1, despite not being able to achieve native sharpness. What we have with DLSS 2 is even better than that IMO, native sharpness is achieved despite few blemishes with the added benefit of a large fps boost.
The problem is they seem to be taking it too far far too often, add to that all the ringing artifacts, overemphasized straight edges etc there's just too many "few blemishes" for me to appreciate it, no matter if there's FPS boost or not. I'm not using naive scaling either, I rather drop settings a notch here and there to get the FPS I want, since IMO the differences between settings like ultra and high or even high and medium are usually far smaller and less noticeable than blurriness with scaling (even with sharpening) or DLSS issues, including 2.0
 
Back
Top