Various Rendering Techniques: Sparse Rendering, Temporal Reprojection, Artifacts Filtering

Just not cluttering https://forum.beyond3d.com/threads/id-buffer-and-dr-fp16.60209/ with this checkerboard discussion.
Funny how they found in "checkerboard" a nicer name for 2:1 resolution Quincunx sampling + up-scale. Wasn't that dismissed/bashed to death? /s

Code:

4x4 pixels with programmed sample positions (will become 4x8 after upsampling), just amplified from 2x2 to actually see Quincunx
.-------.-------.-------.-------.
| *     | *     | *     | *     |
+-------+-------+-------+-------+
|     * |     * |     * |     * |
+-------+-------+-------+-------+
| *     | *     | *     | *     |
+-------+-------+-------+-------+
|     * |     * |     * |     * |
`-------´-------´-------´-------´

 

Isn't one of the benefits of the checkerboard implementations the additional temporal sampling?

 

temporal reconstruction completely changes the game.
And also, in my opinion, a ligh pixel-wide blur doesn't ruin an image so much imo and is borderline unoticeable at 4k.

 

There are examples of great CB rendering at superb fidelity. So no. Quincunx had one solution, blur. CB has many different degrees of implementation giving anything from blurs to checker patterns to crisp, high fidelity renders.

 

results are a mixed bag I think. I noticed incredible results in a game whose name I cant remember (there was a park, and some kind of kiosk with a triangular shaped roof, vegetation in the background and a guy on a skateboard), the comparison screenshot was posted in the forum, and it almost looked like native 4k, but other games like Rise of the Tomb Raider look far worse compared to native.

 

How do you know? Do you have a version that supports CB and full 4k with all other options being the same?

 

Don't like checkerboarding, rather just have traditional upscaling with no artifacts.

On the other hand quincunx actually worked well in games with limited color pallets like Killzone 2. Or even something like Prince of Persia 2008, I found it toned down the very saturated colors and thick outlines nicely.

But yeah, more often than not quincunx had a negative impact on image quality.

 

that game I talked about was running with the same options and checkerboard vs 4k native. As for Tomb Raider, for reference this is how the PS4 Pro version looks compared to the Xbox One version. Looks nice, right?



But what about it compared to the OneX version? Things change, and checkerboard rendering isn't looking that good.



PS4 Pro version compared to original Xbox One. Again, it looks good.



But then, when we compare it to the OneX version the loss of detail in the PS4 Pro version is clear.



Another comparison showing detail loss. Not bad but not as crisp as native 4k. I don't see huge benefits from it.

 

Please @Cyan . You are comparing images with strong DOF (Pro) against images with no DOF (XBX).

 

On GAF some people were thinking the Xbox One X version was better than the PC version and someone disable the DOF and the PC version texture looked as good as the Xbox One X version. If we could disable DOF the texture will look better on PS4 Pro version but less good than Xbox One X and PC version but not by a huge margin like that...

 

You seem to ignore that "checkerboard" is upsampling using Quincunx sample positions. It's a new "hey"-word for something old, because if Sony would have called it "Qincunx upsampling" people would revolted (for no good reason). Qincunx is a sampling pattern, what you make out of it is something else entirely. Everything you can use for Quincunx sparse-sampling could have been used for Quincunx super-sampling in the past. Nvidia used the Quincunx pattern for MSAA in the past (which is not super-sampling), and after research it has been detected that 30° rotated grid sampling pattern is superior. You could get the idea that maybe (try it) rotated grid patterns are also better for upsampling.

Logically:
- Either upsampling with Qincunx is okay (so is down-sampling or MSAA)
- Or 30° rotated grid down-sampling/MSAA is okay (so is rotated grid upsampling)

In conclusion:
- "checkerboard" is the new quincunx, and not better or worst or better and worst at the same time, it's just not any different
- 4k "checkboard" upsampling can be improved upon, because currently it seems Quincunx, which can be improved upon (low hanging fruit improvement)

The reason why Quincunx isn't great is because of the short periodicity of the 45 degree rotated Quincunx grid and the screen-space grid, this leads to moiré in the anti-aliasing (or aliasing in the case of upsampling). Other sampling patterns, like 30° rotated grid, or especially those with the same number of samples per-pixel but varying positions on a larger tile (this has hardware support), have longer periodicity and the resolved result exhibits moiré at an uneven scale which is mostly not detectable by the eye (as a pattern). Eg. stochastic sample patterns (no pixel uses the same sample position inside the pixel' area) would have no periodicity at all (this has no hardware support).

 

how do you demonstrate it is because of DOF? It could well be, apparently. But look at the PS4 Pro vs XB1 screengrabs, I see the same marmalade of blur on the XB1 version than when you compare the PS4 Pro version to the OneX iteration

 

You are wrong on so many levels. Checkerboarding isn't quincunx but dithered (alternating points). There's no group of five samples being enforced. Quincunx was applied in hardware averaging values across five samples, whereas checkerboarding is done in software and you can do all sorts of things with the samples. Importantly, checkerboarding can exhibit dithering, which Quincunx didn't. Ergo it's different. We also have games with impeccable IQ and no blurring like HZD, ergo they are different. Indeed, the discussion that spawned this one had two differing results using CBR - one blurred and one with checkered patterns. The absence of logic in your argument is thus a little shocking - the results are different, and yet you equate the approaches to being the same.

And this explains your argument and logical fallacy. Sony didn't call in Checkerboarding. They didn't invent it. Checkerboard Rendering was first implemented in Rainbow Six Siege. It has sod all to do with Sony other than Sony have implemented a hardware feature to facilitate CB rendering.

I suggest you put your prejudices aside and look at the tech through the clarity of an impartial engineer, because then you'll appreciate the cleverness of it, the potential, and how devs work with and progress the tech across platforms over the coming years.

 

It's obvious... the X version is sharper than that of PC... DOF explains a part of the difference.

Overall, 3 reasons explain the difference :

- Native 4k vs CB
- Higher assets
- DOF







When there is no DOF on PS4 Pro, the difference is more like that :

 

Sony can’t release the ps5 fast enough so we can stop with this checkerboard thing. It’s so inferior to native, it’s not even funny and yes I believe that when the ps5 launches it will become a thing of the past. Up until then we will continue to witness these bizarre arguements about how “great” it is.

 

Not everybody agrees with you when you start poking around in the threads here. People are even looking forward to have better temporal solutions to have more resources to spend on IQ vs resolution.

 

Are you a fan of photo realism?

 

Is it inferior to native? At the moment it's used in cases the GPU cannot sustain 4k with first generation implementations, you are comparing to 4k native using far more GPU resources.

What if we allowed a fairer fight with similar GPU usage, say 5 or 6k checkerboard supersampled down to 4k Vs 4k native. Would that be as cut and dry?