Support NeoGAF

[blu]

But where are you getting that informatoin beyond trying to deduce from what other techniques that are not this one?

What you are saying has a whole lot of implications and assumptions.

What I'm saying is what any signal processing theory would tell you. Doing checkered rendering without alternating the fields at the highest frequency possible (i.e. every frame) means your're throwing away a good amount of perfectly viable information, and that is never a good thing in signal processing.

Start from '3.1 Reverse Reprojection' and go via '3.4 Cache refresh': http://w3.impa.br/~diego/publications/SchEtAl12.pdf

Where if anywhere does it say that the hardware unit compares information from previous frames to generate the new one? All I am seeing is that it is a spatial upscale with just a different method (from where the sample patterns come from).

Nowhere. The hw does not need do much to accelerate this approach, outside of corrections of the partial derivatives for the purposes of correct texture filtering. It's largely software-implemented.

 

[HTupolev]

I'll have to go read the article fully... intuitively I'm not fully getting how that would work out. I shall read the article though!

When they talk about the method, they link to another DF article, which near the bottom has a slide from a recent GDC presentation (and a link to the presentation) showing the filter.

Other DF article:
http://www.eurogamer.net/articles/digitalfoundry-2016-we-built-a-pc-with-playstation-neo-gpu-tech

GDC presentation where the slide came from:
http://alex.vlachos.com/graphics/Alex_Vlachos_Advanced_VR_Rendering_Performance_GDC2016.pdf

 

[Branduil]

The misleading thing about calling it upscaling is that this technique results in true 4K rendering when things are still, since all of the pixels will actually come from rendering in that case. Spatial upscaling can never create additional detail, it just spreads the existing detail over more pixels. In this technique, the additional detail is actually rendered, but it takes two frames to get there. When there is motion, the previous set of pixels will be inaccurate and have to be approximated, and that is more like upscaling (although the checkerboard and temporal aspects will make the imperfections look quite different than with normal, spatial-only, upscaling).

"Uprendering" may be a little better, since people will not automatically believe they understand what it means, while "upscaling" will make them think they understand when they actually don't. But it's hardly a self-explanatory term, either. You'd need to say something like "checkerboard rendering with temporal interpolation/reprojection" to really convey what's happening.

Eh, it results in true 4K rendering in the same sense that early field-rendered PS2 games were 480i. Yeah if things are completely still you get all the same information, but things are almost never still in video games so that's fairly meaningless.

This seems like a pretty good upscaling trick compared to bilinear filtering though, especially at 4K where the artifacts should be less obvious.

 

[foxbeldin]

In a rough, very small nutshell.

Render the red pixels.
Calculate everything else.

Render the red on odd frames, render the green on even ones.

 

[HTupolev]

Render the red on odd frames, render the green on even ones.

The DF article is discussing the spatial upsampling method. But yes, it would make sense to additionally invert the pattern between frames for getting better temporal data.

 

[BONKERS]

You know your hardware isn't up for the job when you have to design an entire piece of hardware just to make sure you can hit decent resolution and make the upscale not look like shit.

R6:Siege already proved that it looks mediocre. Even with MSAA

4k is too early, wasted efforts here.

 

[foxbeldin]

That's not what the article is talking about.

Dunno about the article, but that's how they explained it at GDC for R6S. They render even pixels, then calculate the odd ones with different parameters (color rendered in previous frame, neighboor pixels, velocity etc...) and alternate every frame.

 

[Inuhanyou]

You know your hardware isn't up for the job when you have to design an entire piece of hardware just to make sure you can hit decent resolution and make the upscale not look like shit.

R6:Siege already proved that it looks mediocre. Even with MSAA

4k is too early, wasted efforts here.

So, what your saying is, you just ignored the entire article and made up a fake scenario in your head?

Nice going.

I think your going to have to sit down and admit that the techniques here actually work as intended, otherwise your going to be embarrassed later.

 

[Dictator93]

Dunno about the article, but that's how they explained it at GDC for R6S. They render even pixels, then calculate the odd ones with different parameters (color rendered in previous frame, neighboor pixels, velocity etc...) and alternate every frame.

That is how ubi did something, not necessarily what the hardware scaling is doing.

What I'm saying is what any signal processing theory would tell you. Doing checkered rendering without alternating the fields at the highest frequency possible (i.e. every frame) means your're throwing away a good amount of perfectly viable information, and that is never a good thing in signal processing.

I agree 100% with what you are saying.

Start from '3.1 Reverse Reprojection' and go via '3.4 Cache refresh': http://w3.impa.br/~diego/publications/SchEtAl12.pdf

Thx for the link.

Nowhere. The hw does not need do much to accelerate this approach, outside of corrections of the partial derivatives for the purposes of correct texture filtering. It's largely software-implemented.

This is my point, one that you also agree with. I - without knowing of course - presume that if the upscaling is to be done, and it is to use previous frames, than it is in fact not the hardware unit responsible. Rather an individual implementation in software as part of the engine: thus making the thread title "PS4 Pro provides checkerboard rendering in hardware - no cost to devs" a bit confusing.

The checkboarding would be one free step then in a larger chain of engine design decisions (of many varieties mind you) that actually have a cost in frametime.

 

[gofreak]

That is how ubi did something, not necessarily what the hardware scaling is doing.

I don't think it's a question that an entire approach is implemented in hardware and then that's how it works universally.

By the sounds of it, I think some common bits of functionality in various approaches to this have been abstracted out into custom hardware to relieve the shader pipeline of doing the same. But it's still up to the dev how to arrange things, and tie it in with the more custom work being done on the shader pipeline. So some devs may do purely spatial techniques, some temporal, all mixing and matching their software with the hardware functionality provided in whatever way is optimal for their approach. It's possible there's nothing in hardware specifically catered to the unique aspects of temporal vs spatial super sampling.

edit - yeah, as you've posted above. I don't think it's a matter of some transparent 100% free hand-off. I think it's more a reduced-cost setup. Maybe for some approaches it more or less works out as 'free', for others not quite so much.

 

[HTupolev]

Dunno about the article, but that's how they explained it at GDC for R6S. They render even pixels, then calculate the odd ones with different parameters (color rendered in previous frame, neighboor pixels, velocity etc...) and alternate every frame.

That's how a particular game chose to implement a checkerboard approach, outside of PS4P. It's actually explicitly a bit different; in R6S's case, each square on the checkerboard is one pixel, but the article describes a situation where each square is a 2x2 region.

 

[leeh]

You know you want to say "Uprendered"

Can't we just call it what it actually is, checkerboard rendering rather than trying to tag it another name?

I really want to see the results of this. Can't wait for some DF's on it. Wonder if theres any loss in IQ. Is there any lossless screens out?

 

[gofreak]

Can't we just call it what it actually is, checkerboard rendering rather than trying to tag it another name?

I really want to see the results of this. Can't wait for some DF's on it. Wonder if theres any loss in IQ. Is there any lossless screens out?

There'll be some loss in IQ vs a native 4K render, depending on the technique used and the - for example - the temporal characteristics of the scene.

There should be a big bump in IQ in all cases vs 1080p.

 

[leeh]

There'll be some loss in IQ vs a native 4K render, depending on the technique used and the - for example - the temporal characteristics of the scene.

There should be a big bump in IQ in all cases vs 1080p.

How detrimental though? If you console game on a PC monitor, id expect to see a noticeable difference. Although, if this is a 4K TV in a living room, would it provide enough noticeable difference to actually outweigh the need to spend the huge amount of extra resource to push double pixel count?

 

[BigTnaples]

How detrimental though? If you console game on a PC monitor, id expect to see a noticeable difference. Although, if this is a 4K TV in a living room, would it provide enough noticeable difference to actually outweigh the need to spend the huge amount of extra resource to push double pixel count?

A 4KTV in a living room will provide the exact IQ boost Pro is trying to achieve.

If anything, a PC monitor would hamper the improvement.


If you have a 4KTV in your living room, the Pro is a no brainer.

 

[mario_O]

This is just a fancy name (marketing) for what really is nothing but an upscale technique. Maybe more efficent than other methods, but still just upscaling.

 

[Aceofspades]

Can't we just call it what it actually is, checkerboard rendering rather than trying to tag it another name?

I really want to see the results of this. Can't wait for some DF's on it. Wonder if theres any loss in IQ. Is there any lossless screens out?

Of course there will be loss of IQ, nothing beats native resolution. In case of checkerboard rendering the loss of IQ is apparently less than usual.

 

[Z3M0G]

Is this the trick that Killzone Shadowfall used??

 

[DavidDesu]

It does sound like we have a decent stop gap console to bridge the gap between HD and 4K, and it makes it all the more likely PS5 and Scorpio will be supporting 4K native a lot more than many of us thought they might be able to. Exciting times, and hey for people like me not likely to get a 4K TV for another 2/3 years this still brings image quality improvements to all of our (upcoming) games (and hopefully lots of patches from kind devs for the older ones). The fact this comes at no cost to devs to effectively run at higher res and super sample should mean fairly noticeable IQ improvements across the board.

 

[Dictator93]

Is this the trick that Killzone Shadowfall used??

no.

 

[jelly]

Not a fan if it's that soft look.

 

[gofreak]

This is just a fancy name (marketing) for what really is nothing but an upscale technique. Maybe more efficent than other methods, but still just upscaling.

The fact is that there are a variety of techniques that you can put under the 'upscale' umbrella. Some yield no better information for the final image than the source image allows - some do offer more information, closer to the target res. So I'd disagree with you - the distinctions are there, and go beyond naming.

How detrimental though? If you console game on a PC monitor, id expect to see a noticeable difference. Although, if this is a 4K TV in a living room, would it provide enough noticeable difference to actually outweigh the need to spend the huge amount of extra resource to push double pixel count?

I've no first hand experience, but the impressions from the show - going by the DF article at least - suggest a very big difference between the 4K modes and 1080p modes on a 4K TV display. HDR seems to be as big a deal on top of that also. (I've no idea how far or how close he was to the TV however )

 

[DavidDesu]

Not a fan if it's that soft look.

Soft compared to full native 4K, massively sharper than a 1080p image simply upscale to 4K... it's far better than just raw 1080p, and the results in person sound pretty damn good.

 

[Durante]

Let's remember that everything is relative.

Right, console gamers have very different IQ expectations (particularly per pixel) from many PC gamers (who play on a monitor).

I'm not personally a "sharpness > everything" point of view, but it is common.

 

[NaDannMaGoGo]

Yawn. No upscaling (*replace with your personal favorite term for it*) algorithm will ever come close to native rendering. So, whereas I'll always welcome alternatives and possibly improvements, I'm already sick of people falling for this marketing shit, treating it as though it were something kinda, sorta real 4k.

The fuck is this. Pushing a console as 4k gaming when that console doesn't actually do 4k gaming (yehyeh, some titles can do it native. But are we going to see those regular 'AAA' ones at native 4k and maybe not just 30 but even 60fps? No of course we won't).

 

[blu]

This is my point, one that you also agree with. I - without knowing of course - presume that if the upscaling is to be done, and it is to use previous frames, than it is in fact not the hardware unit responsible. Rather an individual implementation in software as part of the engine: thus making the thread title "PS4 Pro provides checkerboard rendering in hardware - no cost to devs" a bit confusing.

The checkboarding would be one free step then in a larger chain of engine design decisions (of many varieties mind you) that actually have a cost in frametime.

Ah, ok, I see your point now, and I fully agree.

 

[JustenP88]

You know your hardware isn't up for the job when you have to design an entire piece of hardware just to make sure you can hit decent resolution and make the upscale not look like shit.

R6:Siege already proved that it looks mediocre. Even with MSAA

4k is too early, wasted efforts here.

Did you read the DF article? Is that guy an idiot or what? What's the deal?

I keep seeing people who saw the meeting through a compressed stream, if at all, complain about how PS4 Pro is too weak, looks like shit, hardly makes a difference etc... Then, I'm seeing people who saw it in person (DF, Giant Bomb) call the difference the new hardware makes in image quality "revelatory".

They need to get demo units of this thing out asap so all this pedantic concern trolling about "faux k" and shitty upscaling can be put to sleep.

 

[gofreak]

So, whereas I'll always welcome alternatives and possibly improvements, I'm already sick of people falling for this marketing shit

I'm curious what you're referring to, given that they've been remarkably transparent about the subtleties involved here, and open in acknowledging the challenge of reaching native 4K as a matter of course in a console box this year. What they've said is that they can offer a big upgrade that will shine on 4K and HDR TVs... that's about the extent of the most general claims they're making here.

 

[onQ123]

Yawn. No upscaling (*replace with your personal favorite term for it*) algorithm will ever come close to native rendering. So, whereas I'll always welcome alternatives and possibly improvements, I'm already sick of people falling for this marketing shit, treating it as though it were something kinda, sorta real 4k.

The fuck is this. Pushing a console as 4k gaming when that console doesn't actually do 4k gaming (yehyeh, some titles can do it native. But are we going to see those regular 'AAA' ones at native 4k and maybe not just 30 but even 60fps? No of course we won't).

False! if you're able to get better graphics out of using hardware accelerated checkerboard rendering than you can at native 4K than it is better than native 4K.

 

[gundamkyoukai]

I'm curious what you're referring to, given that they've been remarkably transparent about the subtleties involved here, and open in acknowledging the challenge of reaching native 4K as a matter of course in a console box this year. What they've said is that they can offer a big upgrade that will shine on 4K and HDR TVs... that's about the extent of the most general claims they're making here.

Yep if there is one thing is they were very upfront on what they doing when comes to 4K

 

[cheezcake]

False! if you're able to get better graphics out of using hardware accelerated checkerboard rendering than you can at native 4K than it is better than native 4K.

You could say the exact same thing about native 720p and native 1080p. When talking about resolutions just stick to IQ, and the IQ will always be worse in a checkerboard render than native.

 

[jmga]

Comparison between RotTR on PS4 Pro (4k screenshot taken from Gamersyde video) and PC native 4K.

 

[onQ123]

Comparison between RotTR on PS4 Pro (4k screenshot taken from Gamersyde video) and PC native 4K.

The 1st shot is the PC right?

 

[gofreak]

Comparison between RotTR on PS4 Pro (4k screenshot taken from Gamersyde video) and PC native 4K.

The game is native 4K on PS4 Pro I believe.

I guess the PC shot is a direct feed capture, while the other is a video.

 

[Mula]

This seams to be a good way to achive high quality effects and a Sharp 4K image, with
Consoles.

 

[LukasTaves]

It's not upscaling

They both mean rendering at a lower res and constructing a higher res from that data.

The main difference at this point is that usual upscaling takes a final image rendered at low res where any reconstruction technique use framebuffer data, and it's applied during the rendering phase so they can get more clever ways to boost the resolution.

 

[jmga]

The game is native 4K on PS4 Pro I believe.

I guess the PC shot is a direct feed capture, while the other is a video.

This one is supposed to be direct feed.

The 1st shot is the PC right?

Yes.

 

[gofreak]

This one is supposed to be direct feed.

Thanks, maybe someone can pixel count to check D: ?

The most obvious difference is the contrast/colouring looks different, but I'm not sure what that's down to.

 

[jmga]

The PC version of RotTR has a bug with GI on cutscenes IIRC, that would explain the difference in colors.

 

[dr_rus]

Thanks, maybe someone can pixel count to check D: ?

The most obvious difference is the contrast/colouring looks different, but I'm not sure what that's down to.

This difference may be down to settings or a different PP resolution on Pro.

The real difference is in the apparent lack of smaller details, including most of texture details on the PS4P shot. The staircase effect on the high contrast geometry edges is very apparent as well. This is most likely a result of the resolution reconstruction. It is obviously looking better than just pure upscale from some 2560x1440 res but it's obviously not native 4K.

 

[Shin-Ra]

To its credit, Sony understands that a certain level of scepticism will persist here and to that end, several Sony developers on site told us that lead architect Mark Cerny will be going into the specifics of the custom hardware features in the next few weeks.

Wonder if it'll be another lengthy Gamasutra interview.

 

[Dee Dah Dave]

Be interesting to see what the average native resolution of games is before the various techniques come into play to 'hit' 4K.

I would guess 1600p or so with PS4 or slightly better quality graphics.

The end results will be quite nice though. This isn't like upscaling / rejigging 720p. The baseline quality is much better to start with.

 

[pottuvoi]

This is the tech Quantum Break used to get "1080p", correct?

Nope, different method.

Checkerbox as used in Raoinbow Six: the Siege. (Page 44 forward.)
http://www.gdcvault.com/play/1022990/Rendering-Rainbow-Six-Siege

Quantum Break (Page 43 forward)
http://www.gdcvault.com/play/1023870/Towards-Cinematic-Quality-Antialiasing-in

Dust 514 / Ps3 (PAge 26 forward, not sure if they used it in the game.)
http://advances.realtimerendering.com/s2012/CCP/Malan-Dust_514_GI_reflections(Siggraph2012).pptx

Info on MSAA trick
https://mynameismjp.wordpress.com/2015/09/13/programmable-sample-points/

Simply put, there are huge amounts of different methods to get better use of available resources and we will be seeing a lot more of these in future.

 

[M3d10n]

This might make it easier to understand: think of checkerboard rendering as something similar to interlaced video, but using a checkerboard pattern instead of alternating even/odd lines. Only half the pixels are rendered at each frame, so it takes two frames to build a full 4k image. Re-projection techniques are used to reduce loss of detail and either patterns during motion.

Anyway, games using this technique are rendering twice the amount of pixels than base PS4 to start with.

Thanks, maybe someone can pixel count to check D: ?

The most obvious difference is the contrast/colouring looks different, but I'm not sure what that's down to.

Just like interlaced video, if there's no motion it should look identical to native 4k.

 

[gofreak]

This difference may be down to settings or a different PP resolution on Pro.

The real difference is in the apparent lack of smaller details, including most of texture details on the PS4P shot. The staircase effect on the high contrast geometry edges is very apparent as well. This is most likely a result of the resolution reconstruction. It is obviously looking better than just pure upscale from some 2560x1440 res but it's obviously not native 4K.

Not sure I'd call it stair stepping, but I do see a different edge on the Pro image on the ring finger vs the PC image. There's actually a stronger stair-stepping on the PC shot there, the PS4 artifact seems more like a 'feathering'. That might point to reconstruction vs regular sample aliasing.

Not totally sure about the other differences though. I don't have a keen enough eye to be sure what in the texture detail can be contributed to contrast vs other diffs. I'm sure someone here or on B3D can pick apart the differences and their causes.

Whatever the cause, in that still at least I think the results are good. Whatever edge differences there are there seem like a toss-up compared to the colouring differences at least, which I guess is a bug or difference in capture.

edit - the individual hairs are more notably different to the right of her face, though I think the PC shot has more hair strands going on in general. I think it's a keener difference wrt sampling differences however, than the fingers for example.

 

[pottuvoi]

That is how ubi did something, not necessarily what the hardware scaling is doing.

Quite sure that this is not part of hardware scaling.

Most likely Polaris has some tweaks/instructions that make the implementation more efficient, but there is no reason to believe that there would be hardware block to do this.

 

[eso76]

The most obvious difference is the contrast/colouring looks different, but I'm not sure what that's down to.

HDR!

 

[Garrett Hawke]

This one is supposed to be direct feed.

Yes.

it is direct feed but it is still slightly compressed. comparing it with uncompressed pc png's isn't all that useful for image quality and colour.

but we can see that the PC is using a higher setting on PureHair, it has more strands.

 

[pottuvoi]

HDR!

Could be if the source was HDR and then tonemapped to old good monitors.

 

[Durante]

This difference may be down to settings or a different PP resolution on Pro.

The real difference is in the apparent lack of smaller details, including most of texture details on the PS4P shot. The staircase effect on the high contrast geometry edges is very apparent as well. This is most likely a result of the resolution reconstruction. It is obviously looking better than just pure upscale from some 2560x1440 res but it's obviously not native 4K.

Yeah, as you would expect it's most obvious on the high-frequency specular highlights.