2017 (Q4) Bandwidth/Refresh Summit


#24

What do you mean by async mode and to which specs do you refer?


#26

I have read the comments and thank you for your attention to our products. Today we are talking with the software team about the refresh rate issues. Now, Pimax 8K refresh rate is stable at 80Hz and the software team is trying various schemes to debug 85Hz and 90Hz. I will update the refresh rate for any new progress. Thank you


#27

Ok, sorry about async vs sync, I understand that we don’t need async on 8K since we have 2 screens and CLPL is working on panel, no shutters needed.
But 90Hz/ eye is in Brainwarp mode, right? Input is 60Hz/HMD.
Also have question about DSC on MIPI on V1-V2 vs no DSC on V3, is that maybe reason for drop of refresh rate? Panel Driver IC is spec’d at 90Hz with DSC(on MIPI) or no?


#28

[quote=“matthew.xu, post:26, topic:4150, full:true”]
Today we are talking with the software team about the refresh rate issues. Now, Pimax 8K refresh rate is stable at 80Hz and the software team is trying various schemes to debug 85Hz and 90Hz. [/quote]
Thank you, Matthew, for the update.


#29

if you look at some of the steamvr pimax reviews where they show the output it states 80hz as the output as you can see here :slight_smile:
EDIT: This is for the 8kX … but ive also seen one for the 8k where it states that it runs at 75hz. Will look for it!


#30

I think this is in fact the 8k not the 8kx. Pimax have not demonstrated it anywhere.


#31

Good news guys, I found the actual input res. Pimax 8K V1 and V2 use DSC 2:1 with 16b/pixel color 4:2:0 on input. And were stable 90Hz.

If we use same mode on V3, we run into 720MHz limit that ANX7530 has.

Since Driver IC DSC mode is 3:1(2 mipi/panel/IC), we need to send to each panel 3840x1080@90 data, this means actual input res per HMD is 3840x2160. Actual 4K, as Pimax was saying in FAQ on KS, they removed that I think.

Now here is why you can’t have 90Hz stable with such mode of operation:
3840x2160 @ 90.000 Hz Reduced Blank (CVT) field rate 90.000 Hz; hsync: 202.860 kHz; pclk: 795.21 "high risk they say, I say impossible"
3840x2160 @ 80.000 Hz Reduced Blank (CVT) field rate 80.000 Hz; hsync: 179.440 kHz; pclk: 703.40 "Stable"
3840x2160 @ 82.000 Hz Reduced Blank (CVT) field rate 82.000 Hz; hsync: 184.090 kHz; pclk: 721.63 "Risky, 720 is limit"

So it is pixel clock after all. But because of no DSC support on input… GPU has to do DSC compression(or at least color compression) for HMD to operate at 90Hz. That is my understanding. Could be wrong, probably not. BTW DSC on MIPI is by using anx7530 DSC engine 3:1, 2160hx3840v fed into it(not by using DSC but simple sub/up sampling) Huge loss of image quality comparing to DSC method to reduce BW on DP/HDMI(GPU-HMD cable).
Here are some specs, not to think I imagine stuff :slight_smile:


#32

ANX7530 is DP interface/splitter with two display support. It is not a scaler. So it receives 2x 25601440 and outputs the same resolution. You may either consider it as two separate 25601440 frames or one frame 5120*1440 which is split. Either way you need to redo your math.


#33

Well my math is correct for V1-V2, for V3 I need to connect some dots still.
Idea was for sure to give 4K DSC signal(1080x3840) to Driver IC(where is integrated DSC decoder/scaler), how it works with ANX7530 I am not sure, it is maybe just signal inversion 2160x3840, so it uses SBS to send to one eye 1080x3840, not 3840x1080. My math is correct, check pixel clocks, same res just horizontal and vertical res are inverted in signal.
Yep, I don’t need to redo anything:
This is for DSC 3:1.

Also read again this:

Makes sense now that they did put 3840216080 number, does it? It confused me before, I thought they connect only one panel to test it at native res, but it is for DSC 99% sure.


#34

Correction is to be made, actual input res is not 3840x2160@80 on V3.
This is resolution for both eyes that HMD receives on DP input.
Here is math for real number for input res/eye,like what is 2160x1200(1080x1200 per eye) in Vive/Rift.

Output res/HMD is 3840x2160@80 (sent through cable)
Render res is: 4096x2657 / eye 8192x2657 total (SteamVR screenshot)

2560h x 1080v per eye/ target res after lens correction output
1080v x 2560h upscale to 1080v x 3840h per eye/ res to send to anx7530.

Real numbers: 2 x 2560 x 1080 , upscaled to 3840 x 1080/eye is signal sent, just inverted 2x1080v x 3840h is actual input that is sent to HMD.

So we have it, Pimax 8K V1-V3, real specs:
input res per eye: 2560x1080 upscaled to 3840x1080 is sent to HMD.
Input res(“8K HMD”): 2x2560x1080* *-upscaled to 2160x3840(Vertical x Horizontal)

Like Pimax 4K was with 1440p input(1080p on launch), this is actually 4K per eye HMD but with 2560x1080(Wide 1080p) input per eye. I saw that number 2560x1080 before, I couldn’t figure it out why.

Upscale is done on PC, Driver IC does DSC decompression, no upscaling.
Should all backers know this, I think yes. Share if you think the same.

Edit: I knew something is wrong with 2657 number. Its not even number, in tech we usually have multiple of 2 numbers(binary world). This is just random number to confuse backers what is actual vertical res target after lens correction is performed. This can be 1080*1.4 for optimal performance, so 1512. More than that is SS(Steam VR slider), diminishing returns kick in, ok lets assume 1512 is 1x SS, 1.5 SS is beneficial 2268 could be actual number in use, who then just put 2657, notice 657, lol almost 567… This was sloppy job, it’s 2017 guys… You want to hide things from internet? :smile:
@moderators no comment I guess?


#35

Your math is not correct at all. You assumed that the output resolution of the interface chip was 38402160, which is wrong regardless the interface, as Pimax clearly stated that the input resolution for Pimax 8K/5K was 2x 25601440, so the output will most likely be the same.

Then you proceeded to calculate pixel clock for this wrong resolution and then concluded that it does not fit into operating range of ANX chip. Even on V1 or V2 you are calculating with the wrong resolution.

It does not uses “signal inversion” nor anything else you are trying to make up. Read the spec. It is just an interface chip, which takes video stream on Display port on one side and outputs it over MIPI on the other side. No scaling, inversions what so ever.

This picture is for Pimax 4K, which uses one 4K panel for both eyes. Pimax 8K/5K use two panels and different architecture. Do not confuse the two.

They described the characteristics of one panel in 8K, which is indeed 4K panel and they stated that so far they can reach only 80Hz refresh rate on this panel. There was nothing said about DSC and nothing needed to be said about DSC.

Where did you get this data? Pimax alwas said that the input resolution for one eye/panel is 2560*1440.

The pre-warp render buffer is calculated based on supersampling factor and optical system characteristics. There is nothing wrong about ending up with odd number resolution.

SS factor factors the number of rendered pixels, not horizontal or vertical resolution. Rendering 1000x1000 with SS=1.5 gives you 1225x1225 render buffer resolution.


#36

Why do you waste your time? If you backed it your stuck. When the hmd is released everyone will know the truth and if Pimax has lied they will get smashed and company will probably go under with zero sales. If they haven’t lied and all is good you will have a nice hmd and pimax will enjoy success with a ton of sales.

If I was pimax and I lied I would run with money as your company would be stuffed either way, releasing a bad product or running with money has same ending


#37

You are sponsored by Pimax, 99%(waiting for confirmation :slight_smile: ).
I can’t prove anything to you. There is no separate upscaling(dedicated scaler) on the HMD, only DSC decoding(which restores res data(decode) not simply upscale(multiply) when fed properly), hard fact.
Upscaling and signal inversion is on PC side. Who knew?
Same was with 4K, panel receive only DSC signal on MIPI, HW limitation, if you feed in native res(no DSC) you get 2160p@30 interleaved at 60Hz. No more discussion with you(in public).

Pimax 4K actual input res:
Driver IC in DSC 2:1 , 1080v x 3840h MIPI to Panel res (1 MIPI port used)
2560x1080 inverted and upscaled to 1080 x 3840

DSC 3:1 - 720v x 3840h MIPI to Panel res(2 MIPI ports used)
1440v x 3840h per panel(2 x 720 x 3840)
2560x1440 inverted and upscaled to 1440v x 3840h

Same principle is on 8K. Don’t wanna accept, you can’t - if you are biased.

Who doesn’t know math now? xD
I have also details what means 75/80/90 per eye Refresh rate. Panel is in async mode, just like in 4K 90Hz async, DIPI is feeding half panel at 60Hz signal, each panel has 2 parts, they have limit of 60Hz each, if hide them(with shutters on 4K, strobe backlight on 8K) you get async 90Hz, but 75Hz MTP. I am just getting started with debunking the Pimax marketing specs.

Edit: for @AttK0 yes, guesswork from data from actual engineer, not marketing data or sales reps claims how 4K or 8K work when they have less idea than me, since no tech knowledge while making claims/speculations about present/future tech or how it does(n’t) work. I write how it could work, what makes sense to me. To sales reps it doesn’t cause its not in line with marketed data. For tech person, probably has.
http://forum.pimaxvr.com/users/PimaxSWD/activity .


#39

I suspected they are demoing 8K X prototype with 2xHDMI as disguise, because 8K X has 2 X DP 1.4 in specs. And it will turn out, they want to get people who tried V1-V2 to have positive opinion, and they mostly have, so backers expected same experience or better with V3-5. Now I debunked with V1-V2 DSC was used on input res, not downscaling that destroys quality comparing to DSC algorithm.
What is even worse, after using downscaling(1440v to 1080v) they use DSC 3:1 on MIPI, from DSC 2:1 on V1-V2. 80Hz is async, as was 90Hz async on V1, no real difference in perceived MTP, and we are focusing on 80 to 90 hz, when real dark horse is input res, and lack of DSC on it.
@jorgenRe I don’t want to write whole formula, BTW Pimax formula is also wrong. Just wait for tech analysis on all specs. It is same MTP with 75Hz as 90Hz if both are async, input is at same Hz. MTP is measured in ms… You just distract me…

Edit: I will put comparison between V1-2 and V3 soon, non biased, tech based, no marketing BS. Also comparison between V3 and Vive. That will be interesting :slight_smile: Hype, anyone?. Or just Pimax marketing team can reply to my tech review of marketed specs vs prototype specs. :confused:


#40

This is mostly just guesswork tbh. To call it a fact and start flaming people based on it is just ludicrous. You answered to almost none of the good points risa2000 brough up about your posts.

You seem to pull out many of the numbers you base your calculations on from thin air. As risa2000 noted, where did you get these numbers? The fact that your calculations arrive at a result or number that you think is meaningful doesn’t mean anything if the numbers you base them on don’t either.

Your research into the subject is very much appreciated but you have absolutely no reason to flame others. If you think you’re correct, then prove it with what you have and make the others understand. If you expect others to be ready to accept your differing opinion, idea or view, you should be ready to accept theirs.

Nobody should need to post something like this on a technical discussion thread. This is a place for objective analysis and discussion. Don’t turn it into something else.


#41

For us, laymen, who do not grasp the technical details engaging in this discussion is pointless. I appreciate the time and passion put into this thread by some knowledgeable users but this should be kept civil at all times, flaming wars help no one and create anxiety for most people. On a side note, I think most of us following this topic would appreciate the opinion of Lonetech as he was one of the main contributors to this thread and could shed some light on certain assertions made lately.


#42

True. All these numbers and analyses are interesting, but there might be assumptions that turn out to be incorrect. I’m not going to get upset over missed specs, as long as Pimax achieves the minimum promise: 2560x1440 (upscaled to 4K) at 75 Hz and FoV of 200 degrees.

What matters is the end result: Is it the best VR headset (currently available) for the price? That is, will the display, sound, and mic be of high quality? Will there be no dead or stuck pixels? Is the headset comfortable? If all of those things are true, I will be satisfied. If they achieve 85 or 90 Hz refresh on top of that, I will be very happy.


#43

@matthew.xu I think it would be helpful if you could please confirm any specs that are known here. Hopefully then some of us will be brave enough to visit this thread again :grinning:


#44
  1. This might be a silly question, but have Analogix been asked whether they have a better chip coming in the near future or what the level of effort to make a better one would be?
  2. Or, who are Analogix’s competitors? Does a better chip already exist that could be used instead of theirs?
  3. Could multiple chips be used, or is the cost too high to do so?
  4. Is there a way this could be designed modularly and then offer us add-on options to improve overall bandwidth - like for extra money basically turn an 8K into an 8K X?

#45

Was asking myself the same question - in short why not just swap to a chip which does what is needed?