DLSS4 Super Resolution and Ray Reconstruction (Transformer, TNN)

[Florin]

HUB analyzed DLSS4 in great details, they say it's simply amazing! DLSS4 Performance looks like DLSS3 Quality/native.

Tim did some great work here. This is starting to compete with DF-level analysis and it's just a major step all around up from some of his previous upscaling reporting.

Image clarity (without sharpening!) and stability in motion are what we need from an upscaler, and the way this is presented is way more useful than just freezing and zooming frames and pointing out artifacts.

 

[Malo]

How important are native resolution benchmarks and performance if DLSS is not only faster but has better image quality as well? Would you ever run a game in native resolution if DLSS Quality looked better?

IMO no? And it's been that way subjectively before DLSS4 and TNN. I've definitely preferred DLSS Quality vs any kind of TAA native rendering since I got my 3090.

 

[DavidGraham]

More DLSS4 vs native comparisons in two more games showing how competent DLSS4 Performance has become in butting head with native.

 

[Florin]

Tim made another DLSS4 video that focuses on 1440p

 

[trinibwoy]

I know the focus for these technologies is to improve IQ at lower performance requirements but it would be nice to see some coverage of oversampling (DSR) too. In my experience DSR + DLSS can produce very crisp IQ at moderate performance cost.

 

[Scott_Arm]

I know the focus for these technologies is to improve IQ at lower performance requirements but it would be nice to see some coverage of oversampling (DSR) too. In my experience DSR + DLSS can produce very crisp IQ at moderate performance cost.

I tried doing this once and I felt like DSR added input lag. I've never seen it tested extensively.

 

[Scott_Arm]

It's kind of funny how mileage varies between users. Tim's conclusion is that DLSS was dicey at 1440p, but I literally use it in every single game and I'd even used DLSS performance in a few games. I pretty much never use "ultra" settings and always adjust so I'm getting about 120 fps at a minimum, and I wonder if that's part of the reason why I don't see as many annoying visual problems. I'm not sure how many frames of history DLSS keeps to sample from, but I really feel like temporal solutions work a lot better if you have a higher frame rate.

 

[homerdog]

It's kind of funny how mileage varies between users. Tim's conclusion is that DLSS was dicey at 1440p, but I literally use it in every single game and I'd even used DLSS performance in a few games. I pretty much never use "ultra" settings and always adjust so I'm getting about 120 fps at a minimum, and I wonder if that's part of the reason why I don't see as many annoying visual problems. I'm not sure how many frames of history DLSS keeps to sample from, but I really feel like temporal solutions work a lot better if you have a higher frame rate.

I find DLSS to be highly effective on my 27in 1440p monitor. Much more than it was on my old 27in 1080p monitor where even DLSS Quality could look a bit dicey.

With DLSS4 I find it more difficult to tell the IQ difference between presets (it always looks pretty good), but I can tell the performance difference easily. So instead of Quality I'm opting for Balanced or even Performance.

 

[trinibwoy]

I'm not sure how many frames of history DLSS keeps to sample from, but I really feel like temporal solutions work a lot better if you have a higher frame rate.

That’s likely true for any temporal solution. Camera perspective and speed matters too.

 

[Completelu Average]

It's kind of funny how mileage varies between users. Tim's conclusion is that DLSS was dicey at 1440p, but I literally use it in every single game and I'd even used DLSS performance in a few games. I pretty much never use "ultra" settings and always adjust so I'm getting about 120 fps at a minimum, and I wonder if that's part of the reason why I don't see as many annoying visual problems. I'm not sure how many frames of history DLSS keeps to sample from, but I really feel like temporal solutions work a lot better if you have a higher frame rate.

These are from my system. RTX 4080 Super running 5120X1440 resolution. You can right-click and open them in new tabs to compare them at that resolution.

Native Resolution with TAAU

DLSS at Quality settings.

Comparing Geralt's hair and the texture or the light leather pieces on his armor, the metal pieces and cord on his crossbow, and the textures on the metal pieces of the sword scabbards it's pretty clear that the DLSS is sharper and more detailed. It's also noticeable on the ropes on the left. In motion you can see some of the tree tops flicker as they move around while the DLSS is more stable.

I am less concerned about frame rates as long as the gameplay is fluid and smooth. I cannot stand hitching or stuttering. That said, I run pretty much every game I play at max settings.

With previous versions of DLSS I ran native resolution whenever possible and only turned on DLSS in a few games with extremely demanding graphics (Cyberpunk and Alan Wake 2 with Path Tracing for example). However, I am finding that the overall image appears sharper with less flickering and sparkling running the DLSS at Quality than at Native with AA in every game I've tried it with. I may end up running everything in DLSS Quality regardless of performance.