Nvidia DLSS 3 antialiasing discussion

[pjbliverpool]

You're not even going to see those artifacts when they're slotted inbetween native looking frames at much higher framerates.

Watching this tech evolve is going to be something I tell ya.

Yeah I do think people are focusing too much in errors in the individual frames when that isn't how the game will be experienced. This is exactly what Alex was saying in the video and is struggling to find the best way to present.

People are still too stuck in the mind set that if you can see an issue in a single frame, it'll be visible in motion, but that's not necessarily the case here. It might be in some situations though so we do need to find a good way to analyse this. One things for sure though - it's going to be based on moving footage and not still frames.

 

[PSman1700]

As for the "still less artifacts than the PS5 version" (I literally took this poster off ignore an hour ago, took them 10 minutes), of course not. As I've shown, Spiderman can have more artifacts than the PC version with just DLSS2, let alone 3.

Again I'm not expecting DLSS3 to be flawless, but that's just yet another troll reply.

The PS5 version has obvious artifacts in its reconstruction modes. Zoom in 400% and you'd find them on every platform doing any kind of reconstruction. Why the personal attacking though? If i dont agree with your view on what platform or version of the game is the better way to experience this one, then counter that and dont go personal and childish with calling people for trollers. Its totally un-needed.

Yeah I do think people are focusing too much in errors in the individual frames when that isn't how the game will be experienced. This is exactly what Alex was saying in the video and is struggling to find the best way to present.

People are still too stuck in the mind set that if you can see an issue in a single frame, it'll be visible in motion, but that's not necessarily the case here. It might be in some situations though so we do need to find a good way to analyse this. One things for sure though - it's going to be based on moving footage and not still frames.

Well, maybe he's going to watch and dissect screenshots instead of actually playing the game using the tech, which is not his intention anyway.

 

[Remij]

Perhaps - which I mentioned regardless. Considering we routinely pixel-peep here wrt zooming into distant objects to determine rendering inconsistencies when discussing different reconstruction methods, I do think it's not out of the question to note a new reconstruction addition might cause the protagonist to occasionally lose a foot, no matter how difficult it might be for you to hypothetically notice.

As for the "still less artifacts than the PS5 version" (I literally took this poster off ignore an hour ago, took them 10 minutes), of course not. As I've shown, Spiderman can have more artifacts than the PC version with just DLSS2, let alone 3.

Again I'm not expecting DLSS3 to be flawless, but that's just yet another troll reply.

View attachment 7111



Potentially! I certainly hope we're going to see DLSS 2 continue to evolve as well though, there's definitely still room for improvement there. I mean there are some hard ceilings here.

Nobody said there's any issue with analyzing the tech... of course if you check frame by frame you'll easily be able to pick them out.. I'm simply saying these aren't things you're likely going to notice when actually playing a game. What you ARE going to notice, is the much smoother animation.

Those artifacts you're talking about in that video you posted... looks more like the same occlusion issue that was causing issues with the buildings while swinging around the city, causing white patches, than anything else.

 

[Flappy Pannus]

The PS5 version has obvious artifacts in its reconstruction modes. Zoom in 400% and you'd find them on every platform doing any kind of reconstruction.

The DLSS3 artifacts don't require any 'zooming', 3 of those shots are not zoomed in at all. The artifacts are not remotely comparable to what IGTI brings on the PS5, which are still less than what the PC can exhibit with DLSS2 - as I've shown. You want to show your actual comparisons, go for it - nobody is stopping you.

Guess dynamic res and checkerboarding are no different than DLSS now, btw? Since "you'll find artifacts on every platform"? Is that your position, or does actual detail matter?

Why the personal attacking though?

2 seconds later:

Well, maybe he's going to watch and dissect screenshots instead of actually playing the game using the tech, which is not his intention anyway.

Unbelievable.

 

[PSman1700]

2 seconds later:

Calling out someone for trolling is a different offense then stating your not going to play this game using DLSS3. Your the aggressor in this to begin with. I never start taking things personal when someone has their nitpicking views to my favourite platform. Instead its better to nitpick aswell, technically, ofcourse. Not personal.

 

[Remij]

Both of you chill out.. it's not that serious. People are allowed to disagree. We can just say our piece and move on. There's always going to be people who are more analytical and critical of things than others. Was the same thing when DLSS first came to market. I dealt with many of those people, and had to suck it up when at times they were proven right, as they did when I was ultimately proven right.

It's still far too early to tell, and much like DLSS super resolution itself... can have varying degrees of success depending on the source material.. so while it's too early to say anything definitively.. I can definitely say that I believe it's already in an incredible position, and will only get better from here on out.

 

[Flappy Pannus]

Nobody said there's any issue with analyzing the tech... of course if you check frame by frame you'll easily be able to pick them out..

That's the entire purpose of the video though - to actually segregate out the frames so we can see the job DLSS3 is doing on its own. Of course this is a 'frame by frame' analysis, it's how we can see how close DLSS3 is getting to the native image. I'm still interested to see how well it does in these narrow confines, even if it's not an accurate representation of the aggregate. Just as I'm interested to see how FSR 2.0 handles an NPC's hand movement against a bush when zoomed in compared to DLSS, even though I may barely notice it in-game.

I came across the tweet btw because Alex qt'd it approvingly on Twitter, as you note he has clearly stated the difficulty in determining the overall impact of DLSS3 and it's potential artifacts as a subset of final presentation. But it's still worthwhile IMO to see exactly what it's trying to do by itself.

I'm simply saying these aren't things you're likely going to notice when actually playing a game. What you ARE going to notice, is the much smoother animation.

Probably? Maybe? Which is why I said as much.

It's one thing though to respond with "You're probably not going to notice" if someone posts a still with artifacts from a game and says "Yikes! This is unplayable!" - it's a little different when it's simply looking at what the tech is doing in a highly focused manner. I mean you're going to to need a script to respond with "You're probably not going to notice!" to 99% of the posts here critiquing image quality, the vast majority of people probably won't notice the differences in platforms, that should be understood beforehand - but as a tech forum it's kind of par for the course to nitpick over this shit. The tech isn't necessarily being 'dismissed' because it's put under a microscope at times.

Those artifacts you're talking about in that video you posted... looks more like the same occlusion issue that was causing issues with the buildings while swinging around the city, causing white patches, than anything else.

Nah doesn't look like those at all really. For one thing it's been fixed with patches weeks ago, and the occlusion issue was the entire geometry was not being rendered, leading to completely white spaces.

Yes, this could definitely improve (which I always mention as a caveat if I pick out anything regarding this, like walking on eggshells around here)!

But DF didn't break into Nvidia's HQ and steal a 4090, they allowed them an embargo exception as Nvidia felt this was representative enough at this time to have them analyze it. So we are. We can get a better picture with more games relatively soon.

 

[Flappy Pannus]

Calling out someone for trolling is a different offense then stating your not going to play this game using DLSS3.

You're right, I am probably not going to run out buy a $2000 cndn GPU in the next week. If that's your criteria for anyone to look at DLSS3 then enjoy your 5 account forum.

I do, however, have a 3060 which I use with DLSS. And I also have a PS5. That's why I can actually make comparisons between the platforms beyond just anecdotal. Since those DLSS3 images have "still less artifacting than the PS5" according to you (note no platform comparison was made in my post to begin with) - so show it.

It's simple, give me an example of the same level of artifacting you get from the PS5 version. You know, from your PS5. The PS5 that you must own, as by your own criteria, you can't analyze the image quality effectively without owning each respective product...right?

Your the aggressor in this to begin with. I never start taking things personal when someone has their nitpicking views to my favourite platform. Instead its better to nitpick aswell, technically, ofcourse. Not personal.

Just FYI, I did see your meltdown posts in the other thread that the mods deleted, so this bullshit isn't working.

 

[Remij]

That's the entire purpose of the video though - to actually segregate out the frames so we can see the job DLSS3 is doing on its own. Of course this is a 'frame by frame' analysis, it's how we can see how close DLSS3 is getting to the native image. I'm still interested to see how well it does in these narrow confines, even if it's not an accurate representation of the aggregate. Just as I'm interested to see how FSR 2.0 handles an NPC's hand movement against a bush when zoomed in compared to DLSS, even though I may barely notice it in-game.

I came across the tweet btw because Alex qt'd it approvingly on Twitter, as you note he has clearly stated the difficulty in determining the overall impact of DLSS3 and it's potential artifacts as a subset of final presentation. But it's still worthwhile IMO to see exactly what it's trying to do by itself.



Probably? Maybe? Which is why I said as much.

It's one thing though to respond with "You're probably not going to notice" if someone posts a still with artifacts from a game and says "Yikes! This is unplayable!" - it's a little different when it's simply looking at what the tech is doing in a highly focused manner. I mean you're going to to need a script to respond with "You're probably not going to notice!" to 99% of the posts here critiquing image quality, the vast majority of people probably won't notice the differences in platforms, that should be understood beforehand - but as a tech forum it's kind of par for the course to nitpick over this shit. The tech isn't necessarily being 'dismissed' because it's put under a microscope at times.



Nah doesn't look like those at all really. For one thing it's been fixed with patches weeks ago, and the occlusion issue was the entire geometry was not being rendered, leading to completely white spaces.

Yes, this could definitely improve (which I always mention as a caveat if I pick out anything regarding this, like walking on eggshells around here)!

But DF didn't break into Nvidia's HQ and steal a 4090, they allowed them an embargo exception as Nvidia felt this was representative enough at this time to have them analyze it. So we are. We can get a better picture with more games relatively soon.

Ugh... no duh that's the point of the video... and then you just agreed that Alex says how difficult it actually is to show the generated frames and their artifacts...

So you're not arguing against what I'm saying... except you're framing your responses to me like as if you are... Stop that.

And no.. there's no probably about it... cmon

I never said what you were doing was invalid... I never said there's no academic reason to study the frames and see what's actually happening and dig into it... Why do you feel like I was attacking you? I didn't quote you at all.. lol You're allowed to post frame by frames... and I'm allowed to say that you wont even notice those things in motion..

Nah doesn't look like those at all really. For one thing it's been fixed with patches weeks ago, and the occlusion issue was the entire geometry was not being rendered, leading to completely white spaces.

I'd say it looks exactly like that. Does it happen with native+TAA?

 

[Flappy Pannus]

Ugh... no duh that's the point of the video... and then you just agreed that Alex says how difficult it actually is to show the generated frames and their artifacts...

So what's the point of saying "Of course if you analyze frame-by-frame of the video" then? The implication there is clear, it's that you feel DLSS3 is being unfairly misrepresented as to what it brings to the table as a whole. If it's 'no duh', then there's no point to even mention that - a caveat which I already gave in the first place before showing the pics.

I'd say it looks exactly like that.

It just doesn't. This is what that occlusion issue looked like, it wasn't rendering parts of the world entirely:



This is the DLSS3 "occlusion issue""

Does it happen with native+TAA?

I'm not sure what you're asking here - do you mean do you get those artifacts without DLSS3 or the earlier occlusion issue only affected DLSS2? It wasn't, it affected native as well. Impossible to test now as it's been patched though.

If you mean do those DLSS3 artifacts happen with native, then no of course not (would have made separating out those DLSS3 frames from the video impossible).

And btw, I'm not 'attacking' you, and I don't feel you're 'attacking' me. Go back and look at my first reply to you, it was short but perhaps confusing as I didn't want to quote a certain poster and so made a reply to their post in the middle of my reply to you. That was directed to them, not you - but bad formatting on my part.

 

[Remij]

So what's the point of saying "Of course if you analyze frame-by-frame of the video" then? The implication there is clear, it's that you feel DLSS3 is being unfairly misrepresented as to what it brings to the table as a whole. If it's 'no duh', then there's no point to even mention that - a caveat which I already gave in the first place before showing the pics.

Well DLSS3 IS unfairly being represented there in that video... when you literally only show the frames it generates and not the other half (or more) of the frames... you're misrepresenting what the user actually sees...

I SIMPLY said that you wouldn't notice those artifacts when they're slotted inbetween native looking frames.. That doesn't mean they don't exist.. that also doesn't mean there's nothing to gain by focusing on the generated frames and improving the tech... I never said any of that.

It just doesn't. This is what that occlusion issue looked like:

View attachment 7113

This is the DLSS3 "occlusion issue""

View attachment 7114




I'm not sure what you're asking here - do you mean do you get those artifacts without DLSS3 or the earlier occlusion issue only affected DLSS2? It wasn't, it affected native as well. Impossible to test now as it's been patched though.

If you mean do those DLSS3 artifacts happen with native, then no of course not (would have made separating out those DLSS3 frames from the video impossible).

I'm not even talking about DLSS3... I'm talking about your "specular" comparison between the reconstruction techniques...

That looks like an occlusion issue.

 

[Flappy Pannus]

Well DLSS3 IS unfairly being represented there in that video... when you literally only show the frames it generates and not the other half (or more) of the frames... you're misrepresenting what the user actually sees...

Ok, so right there is the disagreement we're having, the intent of showing those frames. 'Duh' indeed.

I'm not even talking about DLSS3... I'm talking about your "specular" comparison between the reconstruction techniques...

That looks like an occlusion issue.

It's not. Those are white specular highlights on the marble/stone. As I mention in the videos description, DLSS2 exacerbates these flickering artifacts when other low-res buffers are employed at the same time - sometimes significantly. Due to these effects being lower precision in nature, you can still see them in native TAA and the PS5, just to a far lesser degree. This isn't entirely out of the norm for DLSS mind you as these kinds of things stand out in other games, but it can vary quite a bit. I think Spiderman's reconstruction in particular was very focused with Insomniac's temporal solution and it just doesn't fit that neatly into DLSS as well.

As for the relevance in this thread though: You brought up the previous occlusion artifacting issue as you felt that's what we were seeing in the DLSS3 video stills. They still look nothing alike when they were actually occurring, and they don't resemble the specular flickering that occurs with DLSS now, so I just don't get the point of this tangent.

 

[Remij]

Ok, so right there is the disagreement we're having, the intent of showing those frames. 'Duh' indeed.

No... I have no disagreement with any intent of showing those frames.... Oh my lord...

Please tell me where I say you shouldn't post them or talk about them? PLEASE... I'll wait.

As for the relevance in this thread though: You brought up the previous occlusion artifacting issue as you felt that's what we were seeing in the DLSS3 video stills. They still look nothing alike when they were actually occurring, and they don't resemble the specular flickering that occurs with DLSS now, so I just don't get the point of this tangent.

Another completely wrong misunderstanding by you.. I didn't bring up the previous artifacting issue as what I felt we were seeing in the DLSS3 video.... What did I say??

"Those artifacts you're talking about in that video you posted... looks more like the same occlusion issue that was causing issues with the buildings while swinging around the city, causing white patches, than anything else."

Clearly I'm talking about the specular video you posted... and I even clarified that afterwards.. and here you are again claming that I'm referring to the DLSS3 video. I'm not. I'm referring to the video you made...

 

[Flappy Pannus]

Why nitpick on the version of a game you dont prefer to play,

The thread where those videos originated that I created was literally the thread about Spiderman PC. The videos are about Spiderman PC. This shouldn't be that difficult. They're being brought in here because you brought up Spiderman PS5 as having worse artifacts than DLSS3.

Still waiting on that btw, whenever you're ready to substantiate it.

 

[Remij]

Anyway... nothing wrong with analyzing DLSS3 generated frames. I don't know why anyone who thinks someone who posts on this forum would expect anything else... Nobody said anything about not examining and critiquing the generated frames.

 

[Flappy Pannus]

While the PS5's reconstruction/upscaling is quite good, it just cant match the more advanced technology in both performance and image quality. Sorry man.

Oh yes, DLSS in image quality in most titles is superior to other temporal upscalers on consoles and checkboarding in particular, sure! Mostly.

Those exceptions can crop up in certain cases though, and Spiderman has them. And they suck when they do, which is why I create videos to show them, as I have for other games (Horizon Zero Dawn's artifacts in cauldrons are another example which can exhibit artifacts that the generally inferior checkerboarding solution doesn't, which is especially annoying due to DLSS's drastically superior appearance in most other areas of the game).

I do these as these issues stand out to me when they occur, and in particular when I rarely see them brought up in other venues. See, I can actually compare these things, on account of...actually having the two platforms and the same games on each (nothing screams 'console zealot' than actually buying the same game again on the PC years later, after all!). I'm certainly not exceptional in this regard, many people likely own more than one and can compare each without being accused of being a platform apostate. Most just don't have that much of an emotional attachment to this stuff.

Now in this particular thread though, I was talking about the artifacts specifically with DLSS3 frames. And you were to, hence your reply:

Still less artifacts than the PS5 version, not too bad for a first iteration.

...that is of course ridiculous, which is why you have to pivot to this hand-wave of "Well, as a technology...". Your point was solely about artifacting, and clearly referencing DLSS3 in particular. It was not about "Well, on the whole, DLSS3 brings new things to the table and is a superior technology" which would be largely beside the point, but not really debatable. DLSS3 is definitely more 'advanced' in what it's doing and temporal upscalers cannot do what it does, which is to massively increase performance even with a CPU bottleneck. But in terms of artifacts, it generates more, period. Measured against the benefits it brings they may be extremely minor, yes. But "less artifacts than the PS5 version" is just pure nonsense.

 

[Shifty Geezer]

Anyway... nothing wrong with analyzing DLSS3 generated frames. I don't know why anyone who thinks someone who posts on this forum would expect anything else... Nobody said anything about not examining and critiquing the generated frames.

This is a common interpretation of internet conversation, where a statement read is either assumed in agreement with or an argument against the point against which it is responding. It's good you and Flappy Pannus managed to untangle your wires, and a reminder to us all when discussing to first question if we are reading the others content in correct context.

Both aspects, the value of the technique in stills and in motion, are part of the debate and evaluation of the algorithm.

 

[Flappy Pannus]

There was a well-phrased question into the latest DF Direct that myself and others have had, which is why has the focus on DLSS3 been with Performance Mode DLSS, and would Quality mode make a difference?

Their response was a little unequivocal imo, not really answering the question as to why performance mode was used exclusively (especially as Rich indicated, Quality mode on a 4090 still has enough headroom to reach 120+ fps with DLSS3 in Spiderman). Alex seemed to downplay the difference of DLSS performance mode vs quality in general, the indication I got is that they won't necessarily be testing the effect of different DLSS quality modes with the full review of DLSS3 which I think would be unfortunate.

I get that the DLSS3 artifacts may be very difficult to determine when we're at 100fps, but that's not necessarily the case with the type of artifacts that lower DLSS modes actually amplify in motion. The good news is that DLSS3 is not tied to a particular DLSS mode at least.

I think DLSS quality modes can vary quite a bit with certain artifacts though, simply because the 'goal' of DLSS3 is the highest framerate possible - as Rich seems to indicate - doesn't really negate that. I mean final end quality does matter, or else you wouldn't be advertising those framerates at '4k'. DLSS performance '4k' can look quite a bit different - especially in motion - than DLSS quality '4k'.

 

[Scott_Arm]

@Flappy Pannus There are a lot of scenarios I'd like to see tested. I'd like to see if you can do DLSS Quality as well as DLSS 2 disabled with DLSS frame generation turned on. Still haven't seen it confirmed if you can do frame generation with native frames. I'd like to see comparisons to tv motion smoothing. I'd like to see true 120Hz+ video with frame generation enabled. Then on top of that I'd like to see various types of games tested, especially first person games which would probably be the most demanding since you can move the camera incredibly fast. I'm wondering if there's some case where the current and previous frames are so different that the generated frame is not possible and is abandoned, or if it just pumps out a really bad frame.

 

[trinibwoy]

A Look at the Possible Future of 3D Graphics: How More Real Than Real Can You Get?

With CPUs and GPUs continuing to get more powerful with each new generation, the push for ever more realistic graphics in blockbuster games shows no signs of...

www.techspot.com

Ten years ago, the top GPUs comprised ~3.5 billion transistors and now we're looking at 76 billion nano-sized switches, with 25 times more FP32 calculating power. Is that huge increase in transistor count just for more processing? The simple answer is no. Modern GPUs are dedicating increasingly more die space to units specialized in ray tracing and AI applications. These are the two areas that will dominate how 3D graphics continue to evolve: one to make everything more real, the other to make it actually playable.