Demostrating Pimax 8k has real RGB subpixels with an 4k panel resolution. And explaining the source of its image problems


#162

What you and @LoneTech are saying definitely makes sense when you look at the pictures. There are clearly pixels being wasted/improperly addressed. The numbers on the gauges in DCS seem to make that clear.


#163

Not really. The pixel density is per eye, so in the overlap region, you see superimposed pixels of the same density.


#164

Exactly! The 0s are white blobs without any black in the center. That’s not an artifact of the G/RB subpixels or of the upscaling. Those black pixels are gone; that should not happen for an upscale.


#165

It’s possible that the two panels of the 8K split the sub-pixels in different directions, you would get an illusion of more detail based on the combined images that can’t be seen in individual lens shots. I don’t know if that’s what’s happening, but it seems like something that could be done in theory. Maybe someone should test if there’s a difference between the visibile detail with 1 eye closed. That might also explain Swevivers’ indescribable detail at a distance with the 8K that defies camera shots.


#166

This fits the theory that this is a G/RB panel.

With a G/RB panel we would have half of the (sub)pixel compared to a RBG stripe panel. So the 2.25 factor between 5K and 8K panel would diminish to 1.125. However the 5K uses 9% smaller panels with the same lenses. 1.125/1.09 = 1.032, which means that the 8K has 3.2% more pixel than the 5K. This is pretty that what your results indicate.


#167

I was doing up some interesting reading. Pentiles have an unusual sub pixel pattern that the display controller uses an algorithm to adjust input to match.

Now this could be maybe adapted into the 8k firmware or adjusted in pi rendering as I believe was suggested by @LoneTech.

The 4k model from what I read up on rainbow rgb is also not what one would consider full rgb either.


#168

If you refer to the Rainbow-RGB in the LS055D1SX04, that is full RGB 3840x1440 with a scrambled subpixel order (permutation varies in a 6 line sequence, presumably to break up subpixel lines). By the same class of interpretation, the 8K’s panel is full RGB 3840x1080 or 1920x2160, with a 2 line pattern (with diagonal subpixel lines both ways). And if we could feed it like what it is, we wouldn’t be messing with scalers at all.

I spent a bit of time trying to illustrate what the scaler levels end up doing: https://drive.google.com/open?id=1O-3nnphhvZSs2edVL7uWbgQ41RhcX3-R … and I’m sure there’s plenty of vague stuff I’ll get roasted for. For instance, I’m aware the colour channels don’t line up perfectly and there’s a bit of colour bleed. And you’ll have to back away from the monitor, because the artifacts will be cancelled (or exagerrated) if you scale the image down.


#169

This forum is going mad again. So much speculation and pixelcounting here.

Facts are:

  • Nearly everyone (testers, backers on meetup) said the SDE is less on the 8k
  • Nearly everyone said both are superior than current HMDs on the market
  • Nearly everyone said both are great

Pimax never mentioned a 8k+ anywhere. There is no 8k+. This just a naming thing. When pimax brougt their new secret prototype to the backer meetup it didn’t even have a name. The testers decided to name it 5k+ and pimax liked it. :wink:

At the moment people get very skeptical again. I think it will turn into anther direction again when the first HMDs are shipped and people report on it. This forum is a rollercoaster… :volcano: -> :sunny: -> :volcano: -> :sunny: -> :volcano: -> :sunny:


#170

arrg. I posted in the wrong thread, i think. Dosn’t matter. Works anywhere here :smiley:


#171

What others said (pixel having less subpixels not relevant for SDE as long as it is used/lit). But maybe also because on 8K it is kind of checkerboard pattern which is probably less noticeable than long straight lines pattern of 5K+ SDE. On the picture I notice both equally well but it is probably different in the HMD.


#172

And you’d be hard pressed to get anyone to explain what they mean by SDE, let alone figure out if two people mean the same thing. The SDE on both models is existant, visible, and has different character. Personally I’m rather peeved that the 8K was engineered with needless aliasing and blurring, and ends up having fewer dots than the 5K+. I don’t know about the subpixel pattern of the 5K, but here’s what it works out to with 5K+ and 8K:

Main pixel grid resolution: 2560x1440 vs 3840x2160
Subpixels (dots) factor: 3/3 vs 3/6
PPI boost (panel size / utilization): 9% vs 0% (reference)
Visible dots proportional to: 13.14M vs 12.44M

>>> 2560*1440*3*3/3*1.09**2
13139435.520000001
>>> 3840*2160*3*3/6*1.0**2
12441600.0

It’s a small enough difference to make it reasonable to choose between the two on taste alone. The amazing part is that the 5K+ panel is so much better matched to the optics (but that could be misrepresentation; perhaps the 9% were not linear but area?), and the disappointing part that half the subpixels are just missing on the 8K’s panel.

Back to SDE. The 5K+ panel has obvious horizontal and vertical grid; the 8K panel distracts from it… using obvious diagonal lines. On e.g. Vive, the primary factor in SDE is the low fill ratio; the dots are far too small, and we see wide gaps between them. The 5K+ panel instead has visibly square pixels; you can notice their jagginess. That’s what the 8K panel avoids, but instead has larger gaps (but a lot of diffuse light across them), diagonal lines, and notable blurring. It’s possible to prefer either. Personally I was hoping for something that didn’t have less subpixels than the pentile penalty.

In terms of panel (not image) pixels per degree, we can estimate the 5K+ to 1440/120=12 and the 8K to 2160/1.09/sqrt(2)/120=11.7. Practically identical.


#173

Check at 33:30. Carmack says there aren’t many small 4K panels available on the market at the moment. Phone makers aren’t rushing to develop them, not worth it apparently. Development will have to come from the VR industry.


#174

It is the next checkpoint. With new Steve Jobs and new Zuckerbergs. But there are only 2 of them spinning the wheels every generation.


#175

There is definitely something wrong with the 8K (3840x2160) panels considering the 2560x1440 panels (only 44% of the resolution), looks significantly better.


#176

Pimax 8k:


#177

That’s the best shot of the pixel matrix which has been posted. Thanks!

The RGB subpixels do not form a square, so (to me) it still looks like an individual pixel is either green or red/blue, just like I proposed back when @SweViver posted his pixel matrix shots.


#178

Thats indeed a great shot, much appreciated!


#179

Is it true, the Pimax 8K will have potentially improved image using next gen 7nm cards, e.g. GPU in 2020 because with more GPU power you’ll be able to increase SS to a higher level than is currently possible on a 2080 TI?

If so, that’s one saving grace for the 8K, whose picture is slightly more blurry compared to the 5K+ due to upscaling 1440p, not just panel design difference - upscaling to horizontal 8K presumably using GPU power Vs leaving resolution at native 1440p hence the 5K+ is faster FPS with the same GPU?


#180

No one can tell, it depends on whether they can do some magic with Pitool (e.g. maybe prepare better source image with concrete upscaler/subpixel matrix of 8k in mind). With current implementation (or more precisely the ones testers used) the raw power itself will not do it. Even on maximum possible super-sampling (getting ~1fps) the image still looked sharper on 5k+ compared to 8k as far as I remember. But it will probably improve 8k more than 5k+ (hence why 5k+ has lesser GPU requirements).


#181

I don’t understand if the image quality of the 8k is limited by the scaler, how could an image be improved? I understand post processing, but you do lose something. Turning a raw image into a jpeg will never have the sharpness of the raw image, especially since the jpeg is a smaller format that is then scaled to the original size.