dlss 2.0 needs to be more widely adopted the technology behind it is mighty useful,and I'm not talking about enabling RT.think what a +50% boost to performance in triple A games does to card's longevity. https://cdn.aveine.paris/blog/wp-content/uploads/2019/03/06100707/Screenshot_2-1-768x384-2.jpg

DLSS > NATIVE (as tested in System Shock Demo V1.1) DLSS looks closer to reference 5K image (5120x2880) than the native image. While running twice as fast. That's according to multiple different image quality metrics - SSIM (structural similarity index), RMSE (root mean square error), PSNR (Peak signal-to-noise ratio), SNR and MAE. All you ground inspectors who don't agree: take it to MATH 😀 Reference vs Native http://atomhard.byethost24.com/pub/AA/Test/ssdemo/Reference vs Native.html https://imgsli.com/NDQ5MDI https://abload.de/img/refvnat8jk4t.png https://abload.de/img/refvnatpnijby.png Reference vs DLSS http://atomhard.byethost24.com/pub/AA/Test/ssdemo/Reference vs DLSS.html https://imgsli.com/NDQ5MDM https://abload.de/img/refvdlssklk2h.png https://abload.de/img/refvdlsspsqkj7.png Native vs DLSS http://atomhard.byethost24.com/pub/AA/Test/ssdemo/Native vs DLSS.html https://imgsli.com/NDQ5MDY

Dlss is a feature to be used along side with rt to compensate for crappy fps and shiny surfaces. Why would someone us upscaling if your fps is high enough native?

Undying:

Dlss is a feature to be used along side with rt to compensate for crappy fps and shiny surfaces. Why would someone us upscaling if your fps is high enough native?

Because it looks better and runs faster. While using less power. [youtube=a9KJo9llIpI] 45m40s mark: notice new Ultra Quality mode, not yet implemented, proly using native input resolution. Performance _is Image Quality, because better performance can usually be transformed to better IQ. Like here by using DLSS 5K. Reference vs DLSS 5K http://atomhard.byethost24.com/pub/AA/Test/ssdemo/Reference%20vs%20DLSS%205K.html https://imgsli.com/NDQ5MDc https://abload.de/img/refvdlss2flj0a.png https://abload.de/img/refvdlss2p52k7m.png

Seems to me in spite of yesterday AMD launch, well, some Nvidia mole's got active now. And yes, while static picture with DLSS can look good(and I can imagine, higher resolution is better/hard to spot glitches), then MOVING picture, in game, looks like trash. Saw it with cp2077, thx, no. Makes game with RT playable, but that's it. Auto/quality, doesn't matter.

GREGIX:

then MOVING picture, in game, looks like trash. Saw it with cp2077, thx, no. Makes game with RT playable, but that's it. Auto/quality, doesn't matter.

Yes, you saw it in one game. A game that is already buggy... It must therefore be true for all games 😀 Move on LOL

Control also not looking as sharp compared to native. Like someone said "vaseline smeared" image.

GREGIX:

Seems to me in spite of yesterday AMD launch, well, some Nvidia mole's got active now. And yes, while static picture with DLSS can look good(and I can imagine, higher resolution is better/hard to spot glitches), then MOVING picture, in game, looks like trash. Saw it with cp2077, thx, no. Makes game with RT playable, but that's it. Auto/quality, doesn't matter.

DLSS is far from perfect and can artifact, as any other AA method. But it's not even debatable: Knowing nothing about previous and the next frame, and how to interpolate between them to prevent pixel jumping/crawling - it's Native that is trash in motion. Hence TAA, DLSS and other AA methods with temporal aa comp. Only someone who's completely clueless would bring "in motion" argument against DLSS - motion is DLSS's forte.

DLSS 2.0/2.1 is fantastic. And it will probably only improve more. And adoption has really picked up. AMD is in trouble, me thinks.

If I could ask one thing for the whole 'DLSS is better than native!' argument, it's that we'd stop getting comparisons to TAA as the default. On account of TAA being garbage.

Exodite:

If I could ask one thing for the whole 'DLSS is better than native!' argument, it's that we'd stop getting comparisons to TAA as the default. On account of TAA being garbage.

Previous AMD sycophant asks what about "in motion". You OTOH want raw image with no temporal aa component what-so-ever. You do not care about motion at all. Make up your mind. Here ya go: https://imgsli.com/NDQ5MjQ https://abload.de/thumb/dlsswxkca.png https://abload.de/thumb/natfsks8.png https://abload.de/thumb/refczvj99.png Numbers don't lie: with no AA - It's even worse for "native" : https://abload.de/img/compvzjft.png https://abload.de/img/comppugk8b.png

Noisiv:

Previous AMD sycophant asks what about "in motion".

You didn't come off well from the previous posts, leading with Ad Hominem is worse. On top of forfeiting the argument. *shrug* We're done here, just some more noise to push into the background. For future purpose, you might want to consider that the bad guy can, indeed, be you.

Exodite:

You didn't come off well from the previous posts, leading with Ad Hominem is worse. On top of forfeiting the argument. *shrug* We're done here, just some more noise to push into the background. For future purpose, you might want to consider that the bad guy can, indeed, be you.

I'm sorry. Maybe you should try with Complaints department. Feelings Hurt section. I can help with Numbers and Facts. https://abload.de/img/compvzjft.png --> Native < DLSS (System Shock V1.1), both with and w/o TAA. Q.E.D.

Noisiv:

DLSS is far from perfect and can artifact, as any other AA method. But it's not even debatable: Knowing nothing about previous and the next frame, and how to interpolate between them to prevent pixel jumping/crawling - it's Native that is trash in motion. Hence TAA, DLSS and other AA methods with temporal aa comp. Only someone who's completely clueless would bring "in motion" argument against DLSS - motion is DLSS's forte.

TAA sucks in most games. It's been widely known for a while.

Motion is definitely not DLSS's forte. Anything that moves in Death Stranding gets artifacted into oblivion, like all those tiny floating/flying rocks. They turn into trails because DLSS can't handle them. And some calculation in some test isn't reality. DLSS, implemented as it is, even in the absolute best case scenarios like Death Stranding, CP2077, and Control, is still very clearly inferior to a native image for close to mid range detail. Only for far away thin details like power lines in DS is it actually better than native. Personally I'd like to be allowed to use DLSS... as SS. I don't know why nVidia/devs don't allow it. Right now the DLSS done is at 2.25x below your selected resolution typically. I'd like the option to go above that or even super sample using native res.

Neo Cyrus:

Motion is definitely not DLSS's forte. Anything that moves in Death Stranding gets artifacted into oblivion, like all those tiny floating/flying rocks. They turn into trails because DLSS can't handle them.

Yes. I've seen that in DS and in M&BII. No. It's not that DLSS cant handle them. It's miss-implemented. Motion vectors and DOF issue. Nvidia is aware. Will be dealt with.

Neo Cyrus:

And some calculation in some test isn't reality.

You are free to ignore Objective Image comparison. SSIM & company are only standards in image quality analysis. There are no ghosting issues with SS that i can see. OTOH I can clearly see that unlike native DLSS lacks distinctive pixel crawling. That coupled with objectively better IQ means that in System Shock DLSS is objectively better than native.

Neo Cyrus:

DLSS, implemented as it is, even in the absolute best case scenarios like Death Stranding, CP2077, and Control, is still very clearly inferior to a native image for close to mid range detail. Only for far away thin details like power lines in DS is it actually better than native.

Mostly true. But what do you mean absolute best case scenarios? Wolfenstein, Nioh 2 and SS here are all easily better implementations. CP2077 is a nightmare scenario for any perf-aware AA.

Neo Cyrus:

Personally I'd like to be allowed to use DLSS... as SS. I don't know why nVidia/devs don't allow it. Right now the DLSS done is at 2.25x below your selected resolution typically. I'd like the option to go above that or even super sample using native res.

Ultra Quality mode: To Be Implemented youtube.com/watch?v=a9KJo9llIpI&t=2740s ATM Quality is implemented at 2/3 of output res.

Just seems like an excuse not to optimize the game. I played the demo when it first came out and there is nothing here that warrants how badly the game runs. Although DLSS is a welcome addition because UE4's TAA is complete garbage and badly configured almost 99% of the time into an artifact filled, blurry smeared mess. And it was no exception here, you needed downsampling in the demo previously with TAA to get decent results. But you still had artifacts and the performance was terrible. I second the people asking for a super sampled DLSS option, i'd also like for them to finally expose the sharpening settings to the user as DLSS is often used with a sharpness value that is too high and causes some ugly sharpening artifacts. And objectively, everything i've seen from DLSS 2.0 is superior to native rendering + shitty TAA. The artifacts present from DLSS 2.0 are far less severe than TAA and really it just seems to boil down to the same old "It's not as sharp so therefore it's worse" argument from people who don't understand what a proper ground truth resolve reference would even look like. Shocker: It wouldn't look like a native resolution image with heavy undersampling/aliasing or the artifact filled mess that UE4 is. DLSS get's far closer to a reference ground truth image than any other technique with this level of speed per frame to date.

God damn it...I spoke too soon. There is a DLSS issue in SS demo with graphic assets on ingame monitors shimmering at certain distances and at certain viewports. Not necessarily in motion, but in stills, at specific angles and distances.

MrBonk:

The artifacts present from DLSS 2.0 are far less severe than TAA and really it just seems to boil down to the same old "It's not as sharp so therefore it's worse" argument from people who don't understand what a proper ground truth resolve reference would even look like. Shocker: It wouldn't look like a native resolution image with heavy undersampling/aliasing or the artifact filled mess that UE4 is. DLSS get's far closer to a reference ground truth image than any other technique with this level of speed per frame to date.

Nope they dont get it. Thank god they never saw how much "crisper" an unprocessed movie looks like compared to public version. By now we'd all be blind from the intricate details and catatonic from shimmering.

If I'm being honest, I can usually only tell the difference between DLSS and native when doing an A vs B. If you were to show me a game running purely on DLSS I'd most likely never notice. I'm sure quite a lot of people will, but I'm just not one of them. So to me, personally, DLSS is a blessing.

Noisiv:

No. It's not that DLSS cant handle them. It's miss-implemented. Motion vectors and DOF issue. Nvidia is aware. Will be dealt with.

The same can be said about AA in most games. Specially UE3 games ported from consoles. Often the choice is between an awful TAA implementation and FXAA crap.