8k vs 5k: is "bad" image due to the scaler?

I think the lack of consensus is a problem. I genuinely feel there is no wrong answer as all reviews said both HMDs are excellent and a step up. The reality is that the reviews all stated they preferred the 5K+ but that it was not a straight and easy choice. Some people in here are acting like the 8K is a failure because the 5K is massively better. Yet at no point did any of the reviewers state the 8K was a failure.

The problem for us backers is what happens if we feel we ended up backing the wrong horse.

What if the after changing to a 5K+ the 8K gets an improvement in up scaling due to either software or via a firmware update. Sweviver has stated this has already happened.

I think no matter which one we go with, there will always be an element of “I wonder would I have been happier with the other one”.

3 Likes

Remember, @SweViver has stated that the photos don’t quite capture reality. He said there are no color fringes; those are either slight chromatic aberrations or were introduced by the camera.

None of these shots (which are quite difficult to make) are being taken by a professional setup, so don’t put too much emphasis on the individual details / imperfections.

It’s clear that the 8K has a diagonal pixel pattern, which will help mask SDE and alias jaggies, but the result is softer than a standard rectangular pattern, like the 5K+ uses. The black areas are due to pixel gaps: Eg. SDE. Those gaps are very small and hard to see, when wearing the headset (according to the beta testers and the Berlin meetup participants).

1 Like

I share those worries but they should not have impact on the decision which device to choose based on 2 facts.

1 Xunshu clearly stated that the devices are as good as they get + probably minor software improvements

The software may enhance the performance of the scaler, but I doubt that it will be significant since we know the scaler chip is a not editable piece of hardware.

2 There absolutely may be improved versions over time (hardware) of the 8k (like any device 4k, even the vive) but this may take ages and is not reliable to wait for. Basically, it will be a different model by then.

1 Like

This is it in a nutshell, but I would like us to lean on Pimax to commit to this one more time. Martin seems to think some substantive changes are on the way. He indicated he will have another video up today if he can.

2 Likes

@neal_white_iii I understand that shots are difficult to make and may introduce additional artefacts, however I’d expect the camera to be the same in both shots and thus defects to be similar, otherwise there is no point doing comparisons. Pimax trusted the testers to guide us choosing between 8k and 5k+, so it’s normal to try to investigate the results when they look odd.

@noro, scaler may have different scaling modes, like the ones in a TV for example. If it’s not the case it is an important info because it means 8k will never be better than 5k.

1 Like

I think people are thinking failure because (like me) they expected the 8k to look about as good if not better than the 5k, and now the cheaper option appears to be superior. That seems to leave the flagship product in a weird position (despite the consensus that it too is a vive pro beater).

Personally I’m really happy about the 5k+, but I’m sad about the dream of a no-compromise headset (lowestSDE AND virtually same detail as 5k+ in one headset)

The scaler has been confirmed as not fixed as of yet. @SweViver reports latest pitool 76 has improved sharpness somewhat. This suggests the process can be improved. But never as good as native panel input.

The general census of the testers if your not planning on updating gpu even to latest coming offerings the 5k+ would be a better choice due to performance & native res input being a sharper pic.

I am getting both as planned when I pledged. Testing Hellblade on the 4k on a Zotac mini 1080 ti I know will likely need more power as Med settings seems to yeild smoother frames.

If you watch a lot of movies though the 8k might be better choice due to minimal sde.

If the 5k was built with pentile Oled then it would look better except where oled shines like infinite contrast. But the pentile matrix would have made it’s sde look worst than the Vive pro.

1 Like

Sorry I’m a bit confused here. Neither the 8k nor 5k use pentile OLED as far as I’m aware? So while I agree with your statement, I’m not entirely certain how it relates to 8k vs 5k+?

Are you saying that the 5k+ only has less SDE than Vive Pro due to not having pentile OLED? I guess that could be right.

I was just saying I was hoping to get basically the same quality image as the 5k, but with less SDE than the 5k. Maybe my dream was unrealistic in the first place, but I guess I’m still happy because I didn’t realise the 5k could look as good as it apparently does.

If the upscaling algorithm is built to work with a straight horizontal/vertical pixel arrangement, I can imagine it would cause problems with a screen where the pixels use a different arrangement, e.g. rotated 45 degrees.

Moreover, with a rotated arrangement, every second row will be shifted horizontally by half a pixel, which can also cause issues. A straight vertical line wouldn’t be rendered as straight in these conditions.

The second problem could be addressed by rendering to a buffer sized so that after the lens distortion transform the image is 5120 x 1440. Then for each even row average pixels two by two, and for each odd row do the same but with an initial offset of one pixel (skip the first pixel in the row). Then send that to the HMD.

Maybe even the first problem should be fixable. Since upscaling is really just averaging adjacent pixels, basically doing additions, it should be possible to apply a compensation so that the “faulty” upscaling produces the correct result in the end.

2 Likes

The trouble is that the fact we are scaling 1.5 in both the ‘x’ and ‘y’ axes you are taking 4 input pixels and going over a 9 pixel area:

o a o
b o c
o d o

Where {a,b,c,d} is the input and o is an artificial output of the upscaler.

My gut instinct says that the above arrangement is optimal for a 3x3 grid, as any other pixel arrangement will result in some pixels having only 1 neighbour.

The problem would come with how the neighbouring ‘3x3’ pixels are generated:

o  a1 o  o  a2 o  ...
b1 o  c1 b2 o  c2 ...
o  d1 o  o  d2 o  ...
o  a3 o  o  a4 o  ...
b3 o  c3 b4 o  c4 ...
o  d3 o  o  d4 o  ...

Results in repeated 2x2 blocks of 'o’s which means that there is only the information from 2 neighbouring pixels (partial information ‘might’ be gatherable from diagonal neighbours but those pixels are further away.

The question then becomes whether it is worth ‘sliding’ some elements to change the 2x2 block into the awkward tetris shape? My gut instinct is that such a change would reduce maximal sub pixel distance from a source pixel:

o  a1 o  o  d4 o
b1 o  c1 o  a2 o
o  d1 o  b2 o  c2
o  a3 o  o  d2 o
b3 o  c3 o  a4 o
o  d3 o  b4 o c4

Where the ‘2nd’ 3x3 column has be shifted down 1 panel pixel (and ‘d4’ is used simply to demonstrate this slide, not intended to actually take that pixel value). I’m unsure whether the 3rd such column should either shift down again (keeping a constant awkward tetris shape), or shift back up (causing alternate awkward tetris shapes). My gut instinct here is that an alternating pattern will repeat every 6 pixels, whereas the constant downshift will repeat every 9 pixels, so constant downshift may result in a less discernible pattern.

Note that a ‘shifted’ pattern returns a greater value if you:

  1. Generate a spreadsheet with 0 for the 'o’s and 1 for the {a,b,c,d}. for 12x12 grid
  2. Below the spreadsheet and starting in the middle create a 6x6 grid with all cells having a formula such that =Max(sum(cells adjacent),4*cell) which would return either:
    a) 4 if the cell is an input pixel, so is fixed.
    b) 4 if the cell has information on all neighbours so has as much information as possible for the scaler to assign a value.
    c) 0-3 depending on how many input neighbours the pixel has if it has fewer than all neighbours.
  3. Sum this formula over all the possible pixels in the arrangement.

This is because in the slid arrangement, the same number of cells return 4, but some of the cells that would return 2 now return 3 (but not all).

None of the above applies if the pixels aren’t arranged in a ‘square’ grid manner.

1 Like

I was thinking after seeing:

There are problems is the scaler that’s for sure, but maybe it’s also due to the fact that the update of the frames on the screen is false 80 Hz, that is, half of the pixels of the odd lines are updated and the pairs are updated after the next scan, in this way they reached 80 Hz, so in the video it looks good because it can not be distinguished but the photos capture a specific moment with half of the outdated pixels.

It’s just an idea.

Maybe it is not the screen or the climber is what is tricked to reach 90hz only recalculates one screen and the other no or maybe the scaler is only badly programmed but I still think it may be a failure of these.

1 Like

@LadenSwallow,
yes it looks like the 1.5 scaling is making things complicated! The problem with buffers is that it takes memory and add latency to the scaler. The ANX7530, if that’s what is inside the 8k is probably not able to do such complex treatments.

@Nowcry,
That’s an interesting theory! That would explain why @SweViver say it does not look that bad in reality!

1 Like

Your dream wasn’t unrealistic. You just didn’t realize it ended in “-X”.

2 Likes

X would be higher quality image rather than the same. True, that would be the ultimate dream if I could power that, but I don’t think that’s going to be possible for a long time. Which is why I changed my 8k-X order to just regular 8k.

Unless you’re telling me that running the 8kx rendered at lower than native would be better than upscaled 8k? If so, I guess we’ll find out next year.

1 Like

Maybe so. At the very least I would expect the low rendered 8KX to look better than the 5K+ (ignoring the diagonal pixel pattern).

On another note, most of the optimizations that are being made for the 8K will probably improve the 8K-X as well.

1 Like

At the very least? But according to the reviews the 5k+ generally looks better than the 8k

Well, the main reason people seem to be in favor of the 5K+ is that it is crisper and performs better. The 8K-X should be even more crisp than the 5K+ because of its higher native resolution. The only visual issue that might pop up is the diagonal pixel pattern.

For performance, I would expect an undersampled 8K-X to be at or below the GPU cost of the oversampled 8K. Even if you had to turn quality settings down I think people would prefer the SDE/crispness improvements of the 8K-X to the 5K+.

2 Likes

Count me as one of those people.

3 Likes

I almost pledged the 8K-X myself. If I had known they would provide loaner units I definitely would have taken the pledge.

2 Likes