Pimax Crystal - status, updates and fixes (Part 2)

Pimax Hardware Pimax Crystal
683

I think the 12k would be a huge sucsess if the company just goes with what worked about 8K-X and what works with the Crystal. Dont shoot for insane specs, just polish and perfect what you know.

I think a 12k that could hit the 8K-Xs “Normal” Fov setting, and have that FOV distortion free, while leverageing the extra resolution of the 12ks 6k panels for 35-49 PPD would be absolutely killer.

Set a balance of features that you have already demonstrated can be achieved, but do it better, with that added bit of polish.

49 PPD would put 12k at the same HD resolution and clarity as a ROKID or HMZ T1 has, but imagine that kind of clarity and image definitiom over 150 horizontal instead of 45 degrees, and without distortion.

The biggest issue that any of the players in wide FOV HMDs has had is in biting off mote thsn they could chew.

The OG KS 8K was Pimax biting off more than they could chrew.

8KX was a massive improvement.

12K should just be getting everythinh that was right with 8K X, plus everything right about Crystal.

3 Likes
684

For PCVR, I honestly do not see how even a 4090 and 7800X3D with 32GB of ram could run the 12k to do its capability justice. Let alone such a PC struggles with just the Crystal in trying to push such high frames with clarity at native resolutions.

Perhaps possible with VR tech and GPU driver/feature support that I’m not aware of, but so far I’m hearing that even people with the above system will be struggling to get the most out of their Crystal.

I mean, to me, ok for the Crystal (If we can get decent FoV lenses) and it’ll only get better once GPUs become stronger and can support VR better.

12K just seems overkill and unless I’m in some alternate universe, no PC now-a-days can really run it to its potential.

I ‘FEEL’ we’re at least 2 years off for the hardware and games to be up to some snuff for the 12K.

2 Likes
685

Running an overclocked 8086K and phantom RTX 4080, I’m GPU bound using my Crystal, it would definitely benefit from RTX 4090 or next generation RTX to realise the potential awaiting in Crystal.

Was previously using 3080Ti and jump to 4080 made big difference

With my Valve index went from 2080Ti to 3080Ti again a big jump forward

The eye tracking and dynamic foveated rendering is providing good performance support in compatible titles.

1 Like
686

I don‘t think anybody, Pimax included, have any illusions about the 12X making any sense if you don‘t utilize foveated rendering heavily. The idea must be to use the eye-tracking for dynamic foveated rendering; but even if that should fail, you can apply simple fixed foveated rendering for a significant portion of the picture because it extends to the very outer range of human vision. I assume you would even hardly notice the low resolution in the range beyond the central 140-150 degrees.

So I think despite the 12K‘s ridiculous nominal resolution, this will never be actually rendered.

In other words, I do see a good chance to make good use of it - provided it is generally a good product.

2 Likes
687

Foveated rendering, AND foveated transport, to mitigate the bandwidth limitations, which is why the notion of a 12k without any logic HMD-side is a non-starter.

2 Likes
688

Not sure what foveated transport is - can you elaborate?

2 Likes
689

It’s just sending the frames to the HMD, with more detail for the bit you are looking at than the peripheral bits, favouring it in whatever data transfer rate “budget” you have to spend.

This does not have to mean only resolution, but could equally well involve e.g. different levels of compression for different parts of the frame…

2 Likes
690

Now this looks interesting!

https://almalence.com/dlplugin

“Digital Lens OpenXR plugin for Pimax Crystal”

Open for testing by Pimax Crystal users

4 Likes
691

That sounds a bit - weird. When applying DFR, the assumption is that the differences is required rendering of the resolution, i.e. 1 rendered pixel per physical pixel in the area of the fovea, and e.g 1 rendered pixel per 4,16, etc. pixels towards the peripheral vision is clearly allocated and will be computed by the GPU and that result is sent back to the headset. So that effectively applies a qualification about which pixel deserves full rendering and where you’d rather bundle them for rendering as one logical pixel.

So all of this data being destined for transmission to the headset should be considered equally important because it the peripheral pixel rendered actually serves to fill 4,16, etc. physical pixels, and if it left out and those pixels show just the value of e.g. neighboring pixel clusters, this missing information hurts just as much as one pixel being wrong in the fovea region.

So all of the resulting rendered pixels should be transmitted at 100% accuracy - that was the whole point of DFR. If the bandwidth doesn‘t support this, it means you should let your PC render a narrower fovea at 1:1, etc… But why would you waste GPU performance on some rendering which then doesn‘t make it to the displays after all? And on top of that you‘d have a second process applying the foveated logic which in itself again will stress the resources (question: which processor would be tackling this? Has to be on the PC side as it happens prior to transmission from the PC).

This seems a bit dubious to me.

2 Likes
692

I should preface by stipulating that foveated rendering and foveated transport can be used either one without the other, or both together. They would use the exact same eyetracking data, when available.

Also, I don’t think am not entirely grocking your intent, but suspect this is mutual. :7

So… Currently, when a game uses e.g. VRS, or quad-views rendering, to render a single pixel for a group of four (or more), either a post-rendering shader fills in the remaining blank three on the bitmap, or the VRS-capable GPU does this on the fly (I don’t know which) – typically by just copying that one pixel to its neighbours within the same cluster; although the filling-in could potentially also be done by some more complex algorithm, from simple interpolation to, say, DLSS.

The bitmap is then sent to the HMD in full.

This means that that one, single rendered pixel is sent to the HMD four times (or more) – the once, plus once more for each unrendered pixel in the same cluster.

With foveated transport, you could send the rendered pixel just once, and let a compositor on the HMD perform the exact same filling-in-the-blanks task the post-rendering shader on the host computer does today, eliminating the traffic of three unnecessarily transferred duplicates/derivatees, and relieving the host of the the filling-in-the-blanks overhead.

Offloading the filling-in to the HMD would probably be more complicated with something like DLLS, since then you would need to transfer motion vectors and z-buffer along with the frames, cancelling out the savings. :stuck_out_tongue:

(EDIT: Even when not using foveated rendering, doing the whole frame 1:1, and then sending it to the HMD by foveated transport, those peripheral pixels are de facto supersampled (assuming quality downsampling), and of higher fidelity than would be their foveated-rendering counterparts.)

3 Likes
693

Correct me if I’m wrong here:

XR2 Gen 2 Platform offers significant improvements in GPU and AI performance, as well as enhanced power efficiency and dedicated hardware processing for tracking, making it a compelling choice for immersive VR experiences

So why isn’t or can’t we get a version of the above chip for a revised version 2 Crystal and even the planned 12K ?

Seems like a more appealing choice and should help with performance, no?

2 Likes
694

Rumoured Meta have a six month exclusive contract on the gen2 chip which is one of the reasons the Pico5 won’t come out till April next year.

Who knows what the minimum order quantity Qualcomm put on these chips. I guess Meta order in the 100s of thousands if not millions.

Due to Crystals price and GPU limitations I guess their ordering power is in the 10s of thousands at best which puts them bottom of the VR pile after Byte Dance, HTC, etc.

This is the problem when a headset gets delayed it’s out of date before it’s released especially when core components are already 2 years old.

I guess all Devs are now moving onto the gen2 variant which leaves todays and future Crystals in no man’s land. Not that a Crystal owner will care when their primary playtime Is on PCVR but I said from the beginning buying a Crystal as a AIO is pointless.

4 Likes
695

It would be even more complicated than that :slight_smile:. What is being transferred to the headset is already distorted (warped) image. The area of the ímage which is the potential suspect for VRS or foveated transport is typically at the periphery of the FOV where there is a) the biggest distortion, b) visible chromatic aberration. So in order to “decompress” the compressed foveated transport and preserve both the colors and geometrical property of the original image, you might need more sophisticated algorithms than what it usually used in pancake industry.

3 Likes
696

True. :7 In the end, I suppose we end up shipping the whole shebang pre-distortion/resampling, without “baking” anything in, in order to let the HMD do it all, and quadruple the traffic instead… :stuck_out_tongue_winking_eye:

1 Like
697

Quest3 v Crystal

3 Likes
698

A little strange that he uses wireless with the quest 3 too compare.

He needed to use a link cable and the right settings in the debug tool,it would have looked a alot better in skyrim…But offcourse the crystal will always wins in picture quality.

2 Likes
699

Okay, understood! Thanks for taking the time to explain (also to @risa2000).

It is complex, this stuff. As Apple have noticed, even good ideas like using bent displays (which means e.g. less barrel distortion, and it fits the face snuggly while producing a nice FoV) bring a plethora of further complications with them. It’s still quite a long road to go.

1 Like
700

Hello @PimaxQuorra :grinning: Keep up the good work.

9 Likes
701

Indeed very nice to see and hear the man behind fixing CS issues. The community is very fortunate to have someone of his caliber and passion.

3 Likes
702

Always my pleasure to be with the community!

8 Likes