Nvidia RTX DLSS/Ray Tracing Discussion

[Trunks0]

You actually wouldn't have wanted to use it in COD:MW as apparently DLSS is glitched. Being worked on though.

https://www.pcgamesn.com/call-of-dut...dlss-aim-drift

After players began to notice the DLSS issues earlier this month, the devs have now confirmed that they are “are aware of an issue where DLSS may impact the accuracy of Weapon optics in Warzone. The team is working on a fix for an upcoming patch.”

[Nunz]

Cold War and MW Warzone are different though.

[Trunks0]

Originally Posted by Nunz

Cold War and MW Warzone are different though.

... damn it!

[Nunz]

Originally Posted by Trunks0

... sob lol

lol

[Ziklitschli]

I like how you finaly seem to understand, and provide a direct answer. but then proceed to drown it out with all that broken record noise.

Originally Posted by Exposed

Does the reconstruction still produce better detail than native when TAA (or equiv.) blurring is out of the equation? -- The answer is still yes.

The answer is no. DLSS reconstruction doesn't produce better detail than native when TAA (or equiv.) blurring is out of the equation etc, at least so far.

Originally Posted by Exposed

This includes comparisons to other forms of AA such as SMAA, MSAA, and SSAA itself.

The problem with you is that from time to time you mix in such outrageous claims, that it's really hard to take you seriously. This, reproducing indistinguishable x64. For what you've just said, 4x4 supersample can be brought that will actually straight improve on detail.

Then you will go but no no, you have to factor in performance, despite the fact that you've just posted a blanket, sweeping, definitive statement.

Then you will cry for proof and analysis. It's this neverending cycle of bs.

Originally Posted by Exposed

Does the reconstruction still produce better detail than native when NO AA is out of the equation? The answer is going to be "at least on par" in terms of fine detail, and "better" since DLSS eliminates all the drawbacks of running with no AA in the first place like razor sharp edges, stepped lines, shimmering, etc..

So straight yes or at least on par.

Yes the final image will be better, because it will be antialiased. That was never in doubt. But it doesn't produce better detail, which was my entire point - to find a way to judge the reconstruction, check its magical abilities, and your OTT claims.

Originally Posted by Exposed

Digital Foundry did a great comparison of a game that basically had no AA. I suggest you watch this and listen to how they bash TAA in the beginning for all it's blurriness and how DLSS 2.0 would perform against this game that had no default TAA to begin with.



Nioh 2's basic rendering lacks much in the way of any form of anti-aliasing at all. It's pretty much as raw as raw can be. So the question is: can DLSS retain its performance advantage and still provide an actual increase to image quality up against native resolution rendering? Remarkably, the answer is yes.

It's a good video, I skimmed through it before. I agree with their conclusion, actually any reconstruction with AA vs no AA will be better overall quality than a no AA picture, especially that for me detail is secondary to antialiasing.

But nowhere in the video did they show, or said, that the reconstruction produces better detail than native. They don't even try that angle, which is logical, as it seems very unlikely atm.

Not to mention, if they saw a potential for it, I think they'd have already shown it long time ago, or suggested.

Originally Posted by Exposed

You are dismissive of their analysis from your own words.

You mean the following quote?

Originally Posted by Ziklitschli

I didn't even watch it heh, I think I managed a part but not sure. I also love them, think they had their hand in making consoles playable, and they're likeable people. But some of the bs from them is too much at times heh.

It was posted together with a video called "Dynamic Resolution a perfect fit for PC gaming". I disagree with the premise, it's bs, dynamic resolution is no good fit for console even, probably the worst cancer in there now. The analysis inside I'm sure is good and honest, but can never overcome the cardinal sin of the title.

Originally Posted by Exposed

They even made a video specifically for your comparison no AA argument, see prior link.

Yes but with slightly different premise. Btw I really liked the description

Originally Posted by Digital Foundry

It's a question we've been mulling over for some time. Does Nvidia DLSS achieve results comparable or even better than native resolution rendering because of the impact of TAA on today's games? What would happen if DLSS was added to a title with minimal anti-aliasing?

and here comes the more important question. If you saw it before, why all this then:

Originally Posted by Exposed

Now you're just being coy. You want to bring up an un-aliased image that almost no one will run because of the excessive razor edge stepping and eyebleeding shimmer just to further your flawed argument against DLSS.

Originally Posted by Exposed

I still don't understand your obessession with a non anti-aliased image. This is a terrible way to play and quite frankly a moving of the goal post

Why would you suggest sidestepping and moving goal posts, when even mighty Digital Foundry is concerned with this?

See my noise comment. This could have been done with 1/10 of the writing and time wasted.

Originally Posted by Exposed

And yet there are other articles from more reputable sites that found the opposite. Including our very own eyes who own Metro Exodus enhanced edition and posted our own comments here in this forum. If you have to "find a guy" to make your point, rather than doing your own homework, then maybe you should rethink how you debate. BTW that guy didn't come up to the same conclusions as you, and his raw DLSS quality screenshots from his folder don't match your conclusions either. I hate to say it you might want to do your own research and provide your own links rather than try to have someone else do it for you and then you skim/summarize over it.

Yes homework, a good work to use. It's you who say things like

Originally Posted by Exposed

DLSS 2.0 resolves fine details and aliasing better than native

Originally Posted by Exposed

DLSS 2.0 will give you finer (actual fine detail) over native

Originally Posted by Exposed

Does the reconstruction still produce better detail than native when TAA (or equiv.) blurring is out of the equation? -- The answer is still yes

so the burden of proof is on you.

And again, the routine. The super optimistic, sweeping claim, appeal to majority, taunts for homework, analysis, research. No.

What does my own research means again? Buying the games and testing them all, or I can't discuss, back to that bs. Yes I read and saw enough of it for the purpose of my case, thank you for your concern.

Yes I'm grateful for helpful members like Trunks0, providing the easy to miss article, thank you man. One of the reasons for a forum, noone with a job or family can read it all.

"The guy" has not shown reconstruction improving on native. I'm not saying it's flawless, or kills the debate.

Originally Posted by Exposed

Detail is directly related to the number of physical pixels. There's no getting around it.

https://i2.wp.com/overmental.com/wp-...02/boixmms.jpg

Ok, so let's reverse. Will 1080p DSR produce better detail than just 1080p on a 1080p screen?

Originally Posted by Exposed

Then prove it. Post your screenshots (I would really rather it be a detailed game and not something old looking like Mechwarrior 5 though). Make sure to include performance numbers.

It's you who came up with how "DLSS is the best antialising on the market", and then I asked if better than sgssaa. Again, burden of proof is on you.

Are you really suggesting that you can prove which AA method is better on screenshots?

Why would you need a game looking better than Mechwarrior to compare antialiasing methods?

Why is Mechwarrior an inferior DLSS 2.0 again, is it game dependable? And if it is, how do you ever make sweeping statements about it.

Why mix in performance again? Please answer, don't break the infinite loop of irrelevant.

[acroig]

Z forgive me but he's been posting lots of proof and you've posted lots of personal opinion on this matter. Also starting a post by calling someone a broken record is making me think you're not discussing this in good faith.

So the way I see it continue the conversation in a civil manner on a subject you clearly know Exposed and you don't agree on or don't, just agree to disagree and move on?

Let's keep the discourse civil please. Thanks.

[Ziklitschli]

He's posted screenshots, I could't see DLSS reconstruction improving on detail there.

He posted a video, but they never say it does either.

So what proof is there? It's the first or maybe second post he kind of shown he understood what I even ask, before tried nothing but to dismiss it.

Now, could I have missed sth on the screenshot? Sure. Then he can point me to it.

[acroig]

I'm just wondering if this is an argument that can have a conclusion rather than escalating into insults.

Carry on but please don't make me close the thread, the horse can't take too much beating anymore.

[Ziklitschli]

You've seen that I wanted to back off when it seemed pointless. But here in what is indeed a sea of the same I can see a drop of we're getting somewhere, now that we established that the comparison to native without TAA actually makes sense, isn't just me in the entire world and not moving goal posts.

I said the broken record noise because I referred to it as noise before. I can back this up easily with few quotes, it's just fallacies on fallacies and straight ignoring what I actually say, and what actually leads to the unpleasant tone. But this got a bit better as well.

I'll refrain from snarky comments from now on. Also, I'm probably 10 times more fed up with this than you, and do not intend to drag it any more than necessary.

[acroig]

Ok Z, thanks for the cooperation, I appreciate it. I have asked Cyko to look on this thread as well.

[Cyko]

Don’t make me get off mah horse

[Ziklitschli]

I'd actually prefer it over TAA in many, many games. Have that choice in Mechwarrior and go for DLSS, not because of performance, it's more pleasant to the eye, TAA is just brutal. FidelityFX on the other hand has that unnatural, really sharpened look.

When it comes to machine learning and all the potential magical result though, I was more inclined to believe in it when it was learning on per game basis, despite the much worse quality back then.

But yes TAA can be really good if the dev isn't just clicking the checkbox in the engine, but works on it.

[Ziklitschli]

Btw is that checkerboarding I catched on the snowy picture? The tree on the right.

Originally Posted by Ziklitschli

Mechwarrior 5 Mercs

[Exposed]

I've read your responses, and your responses to the mods, and I can't help but think there's a level of misunderstanding that could easily be solved by simply expanding your game library and thus experience with DLSS 2.0


As I did before I'm only going to address when I see "meat on the bone":

The answer is no. DLSS reconstruction doesn't produce better detail than native when TAA (or equiv.) blurring is out of the equation etc, at least so far.

False.

Can DLSS retain its performance advantage and still provide an actual increase to image quality up against native resolution rendering? Remarkably, the answer is yes.

DLSS is effectively delivering what all supported titles since Wolfenstein Youngblood have offered: better than native quality in most scenarios. Comparing to a very raw image, we've ruled out the DLSS advantage as being exclusive to TAA presentations - it works for games with next to no AA as well.

https://www.eurogamer.net/articles/d...ring-challenge


However, when putting DLSS up against SMAA without a temporal component, so just regular SMAA, the level of clarity and sharpness DLSS provides is quite similar to the SMAA image. Again, there are advantages here – SMAA does have a fair few remaining jagged edges and some shimmering, which is generally cleaned up with DLSS – but when comparing detail levels we’d say both native 4K and DLSS are similar.

There are some areas where you may notice DLSS actually improves the image quality, such as with some fine patterned areas and other elements with thin lines. This is because Nvidia trains the AI using super sampled images with the clearest possible forms of these details.

https://www.techspot.com/article/1992-nvidia-dlss-2020/

The problem with you is that from time to time you mix in such outrageous claims, that it's really hard to take you seriously. This, reproducing indistinguishable x64. For what you've just said, 4x4 supersample can be brought that will actually straight improve on detail.

Then you will go but no no, you have to factor in performance, despite the fact that you've just posted a blanket, sweeping, definitive statement.

Then you will cry for proof and analysis. It's this neverending cycle of bs.

During the training process, the output image is compared to an offline rendered, ultra-high quality 16K reference image, and the difference is communicated back into the network so that it can continue to learn and improve its results. This process is repeated tens of thousands of times on the supercomputer until the network reliably outputs high quality, high resolution images.

The end result was “anti-aliasing approaching the quality of [64x Super Sampled], whilst avoiding the issues associated with TAA, such as screen-wide blurring, motion-based blur, ghosting and artifacting on transparencies,”

https://www.hardwareheaven.com/nvidi...-ai-rendering/

https://www.anandtech.com/show/15648...motion-vectors

https://legacyvnyl.com/what-is-nvidi...efinition.html

(Though we already know DLSS 2.0 still has minor temporal artifacts present in some cases).

So straight yes or at least on par.

Yes the final image will be better, because it will be antialiased. That was never in doubt. But it doesn't produce better detail, which was my entire point - to find a way to judge the reconstruction, check its magical abilities, and your OTT claims.

See above.

It's a good video, I skimmed through it before. I agree with their conclusion, actually any reconstruction with AA vs no AA will be better overall quality than a no AA picture, especially that for me detail is secondary to antialiasing.

But nowhere in the video did they show, or said, that the reconstruction produces better detail than native. They don't even try that angle, which is logical, as it seems very unlikely atm.

Not to mention, if they saw a potential for it, I think they'd have already shown it long time ago, or suggested.

Then watch their wolfenstein youngblood review where they compare with SMAA. Or their Death Stranding review.

Also, don't go selective editing my quotes just to argue against something I did not even state (which you can avoid by simple quoting the point rather than the sentence, but I suspect this is part of your debate tactic).



Does the reconstruction still produce better detail than native when NO AA is out of the equation? The answer is going to be "at least on par" in terms of fine detail, and "better" since DLSS eliminates all the drawbacks of running with no AA in the first place like razor sharp edges, stepped lines, shimmering, etc..


This describes Noah 2 as I intended and the video backs this up.

Ok, so let's reverse. Will 1080p DSR produce better detail than just 1080p on a 1080p screen?

No, let's not. This is side stepping the point of that HD/4k comparison. Is there is a limit to detail from physical pixels alone? Without a doubt yes, and there is no DSR setting that will provide the same level, or approach the same level of detail as the 4k crop within the same physical space.

It's you who came up with how "DLSS is the best antialising on the market", and then I asked if better than sgssaa. Again, burden of proof is on you.

Are you really suggesting that you can prove which AA method is better on screenshots?

Please don't selectively quote. I said "DLSS is the best antialising on the market for gamers".

There has already been countless comparisons of DLSS against other mainstream anti-aliasing methods be it TAA, SMAA, etc. DLSS is itself an equivalent of high level super sampling so what other methods will give better image quality and performance at the same time? Your argument is basically "Well you're wrong because 100000x Super Sampling is better than DLSS 2.0" while ignoring your framerate would be 100000x slower as well). SGSAA isn't feasible to run at 4k (same hit as DSR), becomes blurrier the higher you go, and has inferior edge aliasing than regular SSAA.

If your statement really had an ounce of merit, it would have been validated by various tech sites eager to dismantle DLSS 2.0 at any opportunity.

Why is Mechwarrior an inferior DLSS 2.0 again, is it game dependable? And if it is, how do you ever make sweeping statements about it.

Why mix in performance again? Please answer, don't break the infinite loop of irrelevant.

Does Mechwarrior 5 use high resolution textures 4k and larger that other games such as Watch Dogs Legion and Cyberpunk uses? Or other high resolution game effects? Isn't your entire argument based on "fine detail" which logically follows the game in question should have "fine detail" to begin with? I see Mechwarrior 5 and see a game that barely looks better than Mechwarrior 4. Not that that it detracts from DLSS 2.0, all the benefits DLSS 2.0 provides is still there, it's just not a good visual showcase from a visual point of view.

As far as performance, that should be plainly obvious. Just ask anyone else who runs DLSS and why they run it.

I'm sorry but all your discussion points are heavily flawed in some way. Notice I had to skip over many of your one sentence replies which were just "noise".

[Ziklitschli]

Ok. As you accuse me all the time of sidestepping, moving goalposts, wiggling out of arguments etc etc (which all is exactly what is the base of your style of posting), and the spotlight for how unpleasant it is is somehow on me, I'd like to take a moment to highlight how you really discuss.

I'll adress the DLSS stuff later, but really want this to be separate, so maybe you will read any of it this time.

Originally Posted by Exposed

No, you don't get to wiggle your way out of the point of that picture or my argument. Is there is a limit to detail from physical pixels alone? Without a doubt yes, and there is no DSR setting that will provide the same level, or approach the same level of detail as the 4k crop within the same physical space.

So this is a good example. Funny you'd say wiggle out.

So let's go back few posts.

This is your post, with the ultimate proof Batman picture, after I told you number of pixels is not a limiting factor for detail for 1080p DSR on 1080p screen.

Originally Posted by Ziklitschli

There's plenty of space on the 1080p side for additional detail.

https://www.nvidia.com/content/dam/e...ep-3-640px.png

It's the averaging process that limits detail on 1080p DSR, not number of pixels. It will at some point ofc, maybe at 8x, maybe at 16, never tested that though.

First thing I said about 1080p DSR is that it looks like stable 2560 detail wise, never said it beats 4k in that regard, just that I prefer stability over detail. You never read or stop to think though, just get hysterical and respond to your own attacks on yourself.

Note how I tell you that it's the way supersampling works that limits detail for 1080p DSR, and post an nvidia picture showing just that, and showing exactly how it will improve detail on the 1080p screen.

Most importantly, note how I remind you, that I said 1080p DSR looks like sth 2560 detail wise, and that I never said it beats 4k in that regard.

So what do you post next?

Originally Posted by Exposed

You even contradict yourself and unwittingly support my point when you do decide to link something, like the DSR picture you posted that only confirms 1080p DSR cannot recreate the level of detail of native 4k just by pixel count alone

Originally Posted by Exposed

You claim 1080p DSR from 4k gives the same level of detail as native 4k. My simple HD/4k picture and your Nvidia image link proves this simply is not the case (despite whatever marketing speak you're reading).

What I do next, gather it all to prove beyond any doubt that I never said DSR from 4k gives the same detail as native 4k

Originally Posted by Ziklitschli

So, as you can clearly see, I never said that it gives the same detail as 4k, actually I did exactly the opposite, denying it time and time again. All I said was that the physical number of pixels is not a limit for 1080p DSR from 4k, so 2x2 supersample. You'd see improved detail from 4x4 still, just like you see improved detail from DSR over plain 1080p, ofc diminishing and I don't know where the limit is. Though if nVidia is doing x64 so 8x8 to teach their little robots, must mean it makes sense still, doesn't it.

So you misrepresent my position, and then repeat it over and over again and this is only one example, you do so much of it that all over the place that it quickly gets impossible to discuss anymore.

What do you post as an answer?

Originally Posted by Exposed

Detail is directly related to the number of physical pixels. There's no getting around it.

[https://i2.wp.com/overmental.com/wp-...02/boixmms.jpg

So, I post this question:

Originally Posted by Ziklitschli

Ok, so let's reverse. Will 1080p DSR produce better detail than just 1080p on a 1080p screen?

What do you post again? Same thing, like I never said anything.

Originally Posted by Exposed

No, you don't get to wiggle your way out of the point of that picture or my argument. Is there is a limit to detail from physical pixels alone? Without a doubt yes, and there is no DSR setting that will provide the same level, or approach the same level of detail as the 4k crop within the same physical space.

Btw I also said multiple times, that there is a limit, just that I don't know where exactly. You can find it in this very post. And that's not what I asked.


So, I can only ask, what is this. I grabbed my head in disbelief 2nd time it happened, but you still do this. It can't be explained with misreading, misunderstanding, IDK.

I never spoke with anyone like you, and never felt that a debate is more pointless. All you need is a lake.

That's only one example amongst many btw, though the most glaring.

[bill dennison]

Can DLSS retain its performance advantage and still provide an actual increase to image quality up against native resolution rendering? Remarkably, the answer is yes.

[Ziklitschli]

Ignoring the "it's because you have 2060" argument, it just speaks for itself.

Originally Posted by Exposed

Can DLSS retain its performance advantage and still provide an actual increase to image quality up against native resolution rendering? Remarkably, the answer is yes.[/i]

DLSS is effectively delivering what all supported titles since Wolfenstein Youngblood have offered: better than native quality in most scenarios. Comparing to a very raw image, we've ruled out the DLSS advantage as being exclusive to TAA presentations - it works for games with next to no AA as well.

https://www.eurogamer.net/articles/d...ring-challenge

So you are quoting an article that is based on video that I told you I watched, and adressed a post before. They talk about generalised "quality", but never mention or show improved detail.

There's nothing in that video/ article that would suggest the reconstruction is improving the detail over native. Why do you post it again, exactly how you did it with the DSR that I highlighted a post before, and what you do with everything. You are debating yourself only.

Originally Posted by Exposed

However, when putting DLSS up against SMAA without a temporal component, so just regular SMAA, the level of clarity and sharpness DLSS provides is quite similar to the SMAA image. Again, there are advantages here – SMAA does have a fair few remaining jagged edges and some shimmering, which is generally cleaned up with DLSS – but when comparing detail levels we’d say both native 4K and DLSS are similar.

There are some areas where you may notice DLSS actually improves the image quality, such as with some fine patterned areas and other elements with thin lines. This is because Nvidia trains the AI using super sampled images with the clearest possible forms of these details.

https://www.techspot.com/article/1992-nvidia-dlss-2020/

Why do you mix those two quotes together? The first paragraph is a comparison vs SMAA, but the second is about the comparison with TSAA T8X.

How you posted it, looks like it gives improved detail in comparison with SMAA without temporal component. I'm not you, so I'll ask if this is intentional.

You'd be all over me already, about how I manipulate, wiggle sth sth.

Originally Posted by Exposed

During the training process, the output image is compared to an offline rendered, ultra-high quality 16K reference image, and the difference is communicated back into the network so that it can continue to learn and improve its results. This process is repeated tens of thousands of times on the supercomputer until the network reliably outputs high quality, high resolution images.

The end result was “anti-aliasing approaching the quality of [64x Super Sampled], whilst avoiding the issues associated with TAA, such as screen-wide blurring, motion-based blur, ghosting and artifacting on transparencies,”

https://www.hardwareheaven.com/nvidi...-ai-rendering/

https://www.anandtech.com/show/15648...motion-vectors

https://legacyvnyl.com/what-is-nvidi...efinition.html

(Though we already know DLSS 2.0 still has minor temporal artifacts present in some cases)

Why are you quoting theory, it was in every article and video ever. Yes in theory it is magic.

It's just spamming so you look like you have sth

Originally Posted by Exposed

Does the reconstruction still produce better detail than native when NO AA is out of the equation? The answer is going to be "at least on par" in terms of fine detail, and "better" since DLSS eliminates all the drawbacks of running with no AA in the first place like razor sharp edges, stepped lines, shimmering, etc..
[/i]

This describes Noah 2 as I intended and the video backs this up.

I actually misread that for AA is out of equation. So yes you can even say better when talking vs TAA, if that's what you mean. Not sure why you bring Nioh, where it's "at best on par", not "at least". But whatever really.

Originally Posted by Exposed

I see Mechwarrior 5 and see a game that barely looks better than Mechwarrior 4. Not that that it detracts from DLSS 2.0, all the benefits DLSS 2.0 provides is still there, it's just not a good visual showcase from a visual point of view.

Yes barely

https://gamefabrique.com/games/mechw...enaries-02.png

vs

[Exposed]

Originally Posted by Ziklitschli

Ok. As you accuse me all the time of sidestepping, moving goalposts, wiggling out of arguments etc etc (which all is exactly what is the base of your style of posting), and the spotlight for how unpleasant it is is somehow on me, I'd like to take a moment to highlight how you really discuss.

I'll adress the DLSS stuff later, but really want this to be separate, so maybe you will read any of it this time.



So this is a good example. Funny you'd say wiggle out.

So let's go back few posts.

This is your post, with the ultimate proof Batman picture, after I told you number of pixels is not a limiting factor for detail for 1080p DSR on 1080p screen.



Note how I tell you that it's the way supersampling works that limits detail for 1080p DSR, and post an nvidia picture showing just that, and showing exactly how it will improve detail on the 1080p screen.

Most importantly, note how I remind you, that I said 1080p DSR looks like sth 2560 detail wise, and that I never said it beats 4k in that regard.

So what do you post next?





What I do next, gather it all to prove beyond any doubt that I never said DSR from 4k gives the same detail as native 4k



What do you post as an answer?



So, I post this question:



What do you post again? Same thing, like I never said anything.



Btw I also said multiple times, that there is a limit, just that I don't know where exactly. You can find it in this very post. And that's not what I asked.


So, I can only ask, what is this. I grabbed my head in disbelief 2nd time it happened, but you still do this. It can't be explained with misreading, misunderstanding, IDK.

I never spoke with anyone like you, and never felt that a debate is more pointless. All you need is a lake.

That's only one example amongst many btw, though the most glaring.

It's great that you had that little conversation with yourself. But you conveniently left out why you went out on that tangent. It's because I (correctly) stated "The reason why 4k DLSS Performance (1080p input) looks so good because it still has 2160p worth of pixels to work with"

to which you responded "1080p DSR also has "2160p of pixels" to work with".

So, rather than accept the limitation that 1080p DSR (from 4k) will always be physically limited to 1080p, unlike 4k Performance which displays full 2160p worth of reconstructed data despite a 1080p input, you decided to go off on a tangent trying to prove a fruitless point regarding unlimited internal rendering while ignoring the physical pixel limitation.

So yes, thank you for dedicating an entire full length post to outline this shortcoming.

[Exposed]

I was looking for "meat on the bone" responses but unfortunately we've reached the point where this was lacking in the entirety. So I will respond as best I can.

Originally Posted by Ziklitschli

Ignoring the "it's because you have 2060" argument, it just speaks for itself.

I don't know what this means.

So you are quoting an article that is based on video that I told you I watched, and adressed a post before. They talk about generalised "quality", but never mention or show improved detail.

There's nothing in that video/ article that would suggest the reconstruction is improving the detail over native. Why do you post it again, exactly how you did it with the DSR that I highlighted a post before, and what you do with everything. You are debating yourself only.

Sigh. I was previously warned from a friend that you can be shown all the evidence in the world but your opinion will never change. And the funny thing is, I don't really care to change anyone's opinions, but summarily dismissing any and all technical analysis or shifting the argument is where I take issue with. This couldn't be more clear with your handling of the Nioh 2 video and my own prior screenshots and every other resource pieced together.

Why do you mix those two quotes together? The first paragraph is a comparison vs SMAA, but the second is about the comparison with TSAA T8X.

How you posted it, looks like it gives improved detail in comparison with SMAA without temporal component. I'm not you, so I'll ask if this is intentional.

You'd be all over me already, about how I manipulate, wiggle sth sth.

Two different paragraphs for two different games from the same link, each with emphasis on detail. Did you not read the article itself? Why are you more focused on how I order the important sections I want you to see rather than the point these sections are emphasizing?

I hate to say this, but this is the type of debate tactics that makes these arguments endless and circular.

Why are you quoting theory, it was in every article and video ever. Yes in theory it is magic.

It's just spamming so you look like you have sth

Ok more summary dismissal of anything presented.

I actually misread that for AA is out of equation. So yes you can even say better when talking vs TAA, if that's what you mean. Not sure why you bring Nioh, where it's "at best on par", not "at least". But whatever really.

Ok....

Yes barely

https://gamefabrique.com/games/mechw...enaries-02.png

vs

I'm sure you're more an avid Mechwarrior 5 gamer compared to the rest of us, but there's a glaring issue here that needs to be said. Since you are gaming on a 2060 and prefer to play in 1080p DSR and the only DLSS 2.0 game you've played is Mechwarrior 5, I don't think you've been exposed to what DLSS 2.0 can truly offer at its best. Absolutely nothing wrong with that and in no way am I disparaging your setup or gaming preference. But when you engage in a deep discussion about DLSS and try to debate more out of opinion rather than experience, you fall into that trap that you've done here. You also insult gamers who do not share your opinion on DLSS who have experienced DLSS first hand in a multitude of games and a multitude of different setups.

[Ziklitschli]

Originally Posted by Exposed

It's great that you had that little conversation with yourself. But you conveniently left out why you went out on that tangent. It's because I (correctly) stated "The reason why 4k DLSS Performance (1080p input) looks so good because it still has 2160p worth of pixels to work with"

to which you responded "1080p DSR also has "2160p of pixels" to work with".

What's most hilarious is I could just add this very quote of yours to further show excatly what I wanted to show there

I left it out for size reasons.

But ok, let's look at the quote again.

Originally Posted by Ziklitschli

1080p DSR also has "2160p of pixels" to work with. Or you mean external. What's with the pixels anyway again.

So with full quote, you can see I meant internal resolution, which is also "pixels".

It is a factual statement, and I don't mention detail.

If I meant "2160 of pixels" as in display, it would mean I'm saying 1080p DSR displays in 2160p.

How does it make sense?

How from all of this you come to conclusion that I say it has the same detail as true 4k and proceed to repeat it multiple times despite telling you over and over I don't?

That after I explicitly told you it's not the case.

Btw another great example of you accusing me of sth (conviniently leave out sth) after doing exactly just that. You're probably just projecting your mentality onto others.

Originally Posted by Exposed

So, rather than accept the limitation that 1080p DSR (from 4k) will always be physically limited to 1080p, unlike 4k Performance which displays full 2160p worth of reconstructed data despite a 1080p input, you decided to go off on a tangent trying to prove a fruitless point regarding unlimited internal rendering while ignoring the physical pixel limitation.

So yes, thank you for dedicating an entire full length post to outline this shortcoming.

Originally Posted by Ziklitschli

Ofc you are pixel limited but less detail is culled when you supersample

Thats from the post you were originally answering. Fits exactly what I was trying to show, which is a bold, continouos strawman the likes of which I don't think I've seen yet.

[Ziklitschli]

Originally Posted by Exposed

I was looking for "meat on the bone" responses but unfortunately we've reached the point where this was lacking in the entirety. So I will respond as best I can.

It took at least 2 walls of text and multiple accusations about moving goalpost, sidesteping, your essays about what you think is my true motive, appeals to authority, majority, inquiries about my setup, before you started posting actual links relevant to my question. I can post another train of those np.

Really, you could just have started with that, could have been nice and swift btw (exactly how I post and operate) but you decided to first try to depreciate my knowledge, setup, intentions, reservations and you still try in this very post (I'm not offended btw, just an observation). It took Digital Foundry and Hardwaretimes asking a similar question to drop that tone and start paying attention.

So, you are really bold, again, to talk about meat on the bone.

Originally Posted by Exposed

I don't know what this means.

Originally Posted by Exposed

Since you are gaming on a 2060 and prefer to play in 1080p DSR and the only DLSS 2.0 game you've played is Mechwarrior 5, I don't think you've been exposed to what DLSS 2.0 can truly offer at its best

Originally Posted by Exposed

I can't help but think there's a level of misunderstanding that could easily be solved by simply expanding your game library and thus experience with DLSS 2.0

It's about the neverending talk about my gpu and games. I have no idea what it would even achieve, make me like it more? Start thinking the reconstruction is producing better detail than native?

Is this a meat on the bone you talk about?

And no, a game that I don't want to play at the moment is not going to sway me towards it. I saw videos and analysis of Control and Metro on DF, I watch almost everything of them, skim through DLSS mostly just to see key points, but watched enough to have a general idea how they look.

If any game is to sway me or make me appreciate it, it's Mechwarrior atm. I laugh a little each time you mention it, as not only it looks good, but also I have a suspicion it might do some things better than other implementations. Look how the lasers are resolved on my pic, I have to check and make a screenshots how it is done on native, it's an uneducated wildguess ofc but maybe it has been designed with DLSS in mind, or sth.

I read this when I fiddled with it https://overclock3d.net/reviews/soft..._mercenaries/1

and while they found some faults, it's a great showcase how it's better than a bad TAA impementations.

Anyway, I told you countless times that it's irrelevant. Meat on the bone, remember?

Originally Posted by Exposed

Sigh. I was previously warned from a friend that you can be shown all the evidence in the world but your opinion will never change.

So now you have a friend with you for the sake of going personal.

Originally Posted by Exposed

Two different paragraphs for two different games from the same link, each with emphasis on detail. Why are you more focused on how I order the important sections I want you to see rather than the point these sections are emphasizing?

This is your qoute:

Originally Posted by Exposed

However, when putting DLSS up against SMAA without a temporal component, so just regular SMAA, the level of clarity and sharpness DLSS provides is quite similar to the SMAA image. Again, there are advantages here – SMAA does have a fair few remaining jagged edges and some shimmering, which is generally cleaned up with DLSS – but when comparing detail levels we’d say both native 4K and DLSS are similar.

There are some areas where you may notice DLSS actually improves the image quality, such as with some fine patterned areas and other elements with thin lines. This is because Nvidia trains the AI using super sampled images with the clearest possible forms of these details.

https://www.techspot.com/article/1992-nvidia-dlss-2020/

Does it look like the 2nd paragraph is a continuation of the first, or not?

Originally Posted by Exposed

Did you not read the article itself?

No, I didn't.

I checked how are they testing, and conclusions relevant to my question.

I'm not here to educate myself about DLSS any more than necessary. I'm not here to check all the screenshots of it in the world. I'm not here to read 3 separate links with theory. I could just as well post a 50 page supersampling paper and demand you educate yourself first. This is unnecessary and unrealistic.

I'm here to ask a simple question. Is the reconstruction producing better detail than plain native. You fish for quotes but noone claims exactly that, and seeing how the tech sites seem to have eye on this, I'm sure we would alredy be informed if that was a case. In fact, I think, we would never hear the end of it.

Which is where we can finish the discussion. I don't see it anywhere, I'm sure if DLSS 3 or 5 does it, we will know. I even expected it to maybe show here or there just by sheer chance with all machine learning, but not so far.

Originally Posted by Exposed

I'm sure you're more an avid Mechwarrior 5 gamer

I'm not, I said I play only instant action. I don't have time for campaign, and it wasn't that good so far to warrant wasting it.

Originally Posted by Exposed

You also insult gamers who do not share your opinion on DLSS who have experienced DLSS first hand in a multitude of games and a multitude of different setups.

By saying how I think it's the best reconstruction on the market by far?

By saying how I prefer it over native 4k with TAA, or even sharpened TAA?

Or by saying I don't believe the reconstruction component is producing better detail that a plain 4k image?

You have a nerve to accuse me of shady discussion tactics, and then post this.

[acroig]

OK guys, this is not getting anywhere so one post each to wrap it up and then we move on, OK?

[Ziklitschli]

OK

[acroig]

Thanks.

[Exposed]

I read this when I fiddled with it https://overclock3d.net/reviews/soft..._mercenaries/1

and while they found some faults, it's a great showcase how it's better than a bad TAA impementations.

The question now is whether or not DLSS generates images that are comparable to native resolution renderings. If they are, Nvidia's DLSS 2.0 will give Geforce RTX users what's effectively a free performance upgrade.


Overall, the image with DLSS enabled appears to be slightly sharper, but overall, the difference in image detail is minor.

The takeaway is for Mechwarrior 5, for this reviewer you linked, is that it is at least on par. Essentially reconstruction is doing it's job as intended, making 4k DLSS indistinguishable from native native while providing all the benefits of AA and increased performance. But you already have this game and already stated you prefer this mode. So I'm not sure what else there is to discuss here.

Originally Posted by Exposed
Did you not read the article itself?

Originally Posted by Ziklitschli
No, I didn't.


I have an issue with this, because if you did read the link I provided, it would have answered all your questions and you would see where the sections I quoted came from.

I'm here to ask a simple question. Is the reconstruction producing better detail than plain native.

Plain native that includes any form anti-aliasing whether if it's SMAA, TXAA, MSAA, etc.. etc? Yes, and this has been validated by countless websites. Not always the case, as the goal of DLSS 2.0 is to provide an indistinguishable image from native 4k. In same cases, it can be better, like in Control and Death Stranding, Cyberpunk, etc...

Plain native that has no AA? Sometimes. Again, the goal of DLSS 2.0 is to provide an indistinguishable image from native 4k. From the Nioh 2 testing, it does exactly that. I also provided my own screenshots with an image panel that clearly shows DLSS 2.0 cleaning up text on a panel vs native 4k that had badly aliased text on the panel. Granted it was the only area of the entire screen where it was clearly visible, but it showed my point didn't it?

I have no intention of replying after this, I've said all I could convey and provided all the analysis I could, whether or not others want to listen.

I've had others thank me for my contributions in the last several days, this thread in general over the last year or so, and I'll continue to contribute as more DLSS news and tech comes along, often with my own findings and screenshot analysis.

DLSS is amazing tech with minor shortcomings and allow gamers to crank up the settings, especially for ray tracing, and provide performance and nearly indistinguishable image quality from native resolution, or in many cases better (yes better). It's why AMD is pursuing their own implementation and is even the future of consoles, the most immediate being the Switch Pro which is expected to use DLSS upscaling technology.

I do hope one day you get to expand your game library and play a few of the newer, more demanding games where DLSS really shines.

[Exposed]

Originally Posted by SIrPauly

Damn, didn't know about the mipmap bias tweak in Nvidia Profile Inspector would improve a DLSS/developer limitation or over site to some degree.

https://www.youtube.com/watch?v=6BwAlN1Rz5I

Thanks Exposed for sharing that video.

edit: really enjoy individuals like Alex that discuss some cons.


This was in the comment section and think the individual is right on target:

I wonder if this is present in any other DLSS enabled game and if so, if the mip map fix corrects it. Would be great if they did a followup.

[Ziklitschli]

So, to abuse my last word position as little as possible, I will only answer what will point to you the eternal issue with the discussion.

Originally Posted by Exposed

The question now is whether or not DLSS generates images that are comparable to native resolution renderings. If they are, Nvidia's DLSS 2.0 will give Geforce RTX users what's effectively a free performance upgrade.


Overall, the image with DLSS enabled appears to be slightly sharper, but overall, the difference in image detail is minor.

The takeaway is for Mechwarrior 5, for this reviewer you linked, is that it is at least on par. Essentially reconstruction is doing it's job as intended, making 4k DLSS indistinguishable from native native while providing all the benefits of AA and increased performance. But you already have this game and already stated you prefer this mode. So I'm not sure what else there is to discuss here.

Yes, no discussion basicaly, I just posted it because I think it's as good a showcase of DLSS as any, with a terrible TAA that makes DLSS an obvious choice the very second you switch to it.

But then you use what might be the best example of a problem with testing against TAA, to point to native, and even draw conclusions about reconstruction. It's like I never said anything, this semantic hurdle that seems to grow every time I try to jump over it, and it's not that I didn't try.

Look at what TAA does to the image there btw. It's unbelievable, I never had issue with AA blurring my game, I always put antialiasing before detail, but this is just smear and a type of it that somehow makes even a 4k image straight disgusting. Even when I DSRed it on 1080p screen, which ussualy helps, I still went for DLSS.

Originally Posted by Exposed

I also provided my own screenshots with an image panel that clearly shows DLSS 2.0 cleaning up text on a panel vs native 4k that had badly aliased text on the panel. Granted it was the only area of the entire screen where it was clearly visible, but it showed my point didn't it?

Originally Posted by Ziklitschli

I told you it's about judging the reconstruction, and nowhere on those pics is DLSS reconstruction doing a better job than native.

Originally Posted by Ziklitschli

I could't see DLSS reconstruction improving on detail there

Again this is ignoring everything I ever said in the thread, an absolute refusal to acknowledge my position, and, to risk a word bit too strong maybe, a form of disdain.

Why, I have no idea. Maybe if you didn't assume my dishonesty or cling so much to waving of credentials, or put an actual thought into it, IDK.

Anyway, everything is said and done, and I accept that you refuse to accept my angle. In a grand scheme of things, this whole debate is a definition of first world problems, and a sign that we're probably both doing quite good atm.

And yes, the DLSS picture is better. The no AA picture is unusable, playable only if there is absolutely no AA to apply. Even then, I'd probably straight drop res a notch just to make the TV scaler smooth it out a bit. So yes.

Originally Posted by Exposed

DLSS is amazing tech with minor shortcomings and allow gamers to crank up the settings, especially for ray tracing, and provide performance and nearly indistinguishable image quality from native resolution, or in many cases better (yes better).

Originally Posted by Exposed

I do hope one day you get to expand your game library and play a few of the newer, more demanding games where DLSS really shines.

Thanks, I sure will, Metro is a good game, 30 fps or not heh. I was close to getting it a few times, and had sth approaching a fix not that long ago. It's actually a title where I wouldn't mind DLSS, because of the visible RT noise. So, no matter how pure, true and stable I'll make it, it will still have sth that bugs me, might as well let go a bit. It's this thing where for example shimmering AO makes me more tolerant of sharpening, because hey, shimmers anyway.

I do hope you get Mechwarrior, not sure I would reccomend it for the current price though. Not every game KAC says is a turd is a turd heh, was sure right for the launch as Epic version was basicaly a paid beta, but it got better, has some mods, looks good for what you'd expect from a sim (well simcade here) and is fun. Not to mention, DLSS 2.0.

[acroig]

Metro Enhanced is a great RT/DLSS showcase. IMHO top 3 RT/DLSS games are Metro, Cyberpunk and Control.

All can be had for cheap, I saw Metro for $8 last week.

[Ziklitschli]

Especially that it's on GOG, which is always a plus for me.

But, approaching 40, I get less and less fixes on games and themes, which is quite sad actually. Haven't even installed Kingdom Come Delieverance, and that's a game I'd ussually just grab some book from the era and disappear for a weekend asap. Hope it returns, though maybe this is required to finaly do sth bigger in life, IDK. I play more of competitive games, but get bored with a title as soon as I start winning on regular basis, how many times can you repeat that cycle of get beaten, get ok, get good anyway. Not a story for this thread I guess though.

Maybe I'll buy on next sale just to see the perf. I think watching the DF video, I came to the conclusion that with DLSS I'd get sth around 30 fps. Which is not bad actually, sometimes getting a cutting edge game and see it destroy your PC adds to the appreciation of the tech and graphics on display, if that makes any sense heh.

[Ziklitschli]

Btw can anyone confirm this is checkerboarding? Tree on the right side.

I always try to take at least one pic during movement, since post AA started to be a thing, and think I catched some.