Pimax 8K Refresh Rate Progress

Pimax 8K Series
1

Dear all,

Sorry for the late response. It was Sunday night in Shanghai.

The Pimax hardware team is focusing on building the new V4 prototype, so we asked our new representatives to handle the NYVR roadshow. The refresh rate feature is mainly related to hardware and firmware, but the reps at SFVR do not have very strong hardware background, their explainations would be insuffient or inaccurate. In fact, It’s their first time to travel to the US.
There’s some misunderstanding because of the language and culture differences, and it’s our fault not to sync up in time with the staffs in NYVR about the new progress. Let’s try to reveal more details regarding the refresh rate here.

Scenario 1: without “brain-warp”

Our HW system supports 90Hz, including the bandwidth of DP \processor\ driver IC. As you know, our original solution is 2HDMI, and then we moved to 1DP. These two solutions are very different.
The games already can run at 90Hz on the original solution (not very stable, we have to spend around 1 hour to set up the demo kit.) The stability of high refresh rate (>85Hz) is still a little bit unstable in the new DP solution in V3 prototype, so we decided to lower the refresh rate to 75Hz for the new public demos.

We are currently conducting tests for 75Hz to 85Hz. The 82Hz is already passed the initial test, now we move on to optimize the prototype to 85Hz, and do more compatibility and stability test.

https://youtu.be/ijlzO4L4L8c

90Hz is the upper limit of the whole system. The hardware reasons that may drop the refresh rate: the clock rate, cable length, display adapter FPC, EMI desense issue, Contact impedance from different Video Card’s DP connectors …

The stability is so important because our HMD need to work well with thousands of different computers.
Before we can give a clear answer, we still need to optimize HW design and conduct much more compatibility test.

Scenario 2: with “Brain-warp”

The refresh rate is important to VR experience and it is challenging to system architecture. That’s why Pimax brings out Brain-warp to our users. The technology will definitely double the refresh rate and reduce MTP.

All games can run 120Hz-180Hz as long as your GPU support that. You can choose to enable or disable the Brain-warp function after the performance test when you get the HMD.

If you enable brain warp feature, you don’t need care about the 5Hz or 8Hz drop of the refresh rate.

MTP = senor response time+ GPU rendering time + 1/refresh rate + display switch time
The refresh rate only one of the many factors that influence the MTP. There’re many other solutions to bring out good MTP performance. Including ATW/ASW/BrainWarp/…

We will update the latest risks analysis of the whole project in two days. This will allow our backers do their own assessment again and have a final chance to make the decision before the Kickstarter campaign ends.

Best Regards,
The Pimax Team

16 Likes
2

Hi @deletedpimaxrep1, thank you this is much appreciated.

May I ask, have your team reached a decision on whether or not an inability to hit a consistent 90hz by January would delay shipping or would you manufacture and deliver HMDs with a lower but stable rate?

1 Like
3

@deletedpimaxrep1 It’s good that you talk about the problems and try to solve them.
The main thing that you were open and tried to talk about the problems that people are not worried and didn’t panic

2 Likes
4

We still need more optimization and compatibility tests and will keep updating the community our progress.

4 Likes
5

And here’s the resolution: 4096x2657 on the eye with SS 1.0
The 4K version had 3224x1989 [or 1612x1989 for each eye] by default.
Well, on this resolution, you can already judge the relationship between the parties and the wider FOV. Roughly speaking, in 4K the ratio was 16:20 per eye, but here it was 16:11, which indicates the superiority of the width above the height. And in 4K was the other way around.
A clear difference in the aspect ratio between:

f.png1193×725 2.35 KB

It looks good :slight_smile:

1 Like
6

Why do you have it running at 4096x2557 per eye? I thought the scaling chip of the non X could only handle 2560x1440 per eye? You’re supersampling that much by default?

1 Like
7

Sorry, How did you get the running is 4096*2557?

1 Like
8

Afaik supersampling means that the GPU is rendering with 4096x2557 per eye, applies lens wrapping etc. and then scaling it down to 2560x1440 before sending it to the HMD. If it has this supersampling level at setting x1 then I assume this is the supersampling level that Pimax suggests as viable default setting (as SteamVR will use x1 if the user doesn’t change it - guess 90% of the people will never touch this setting)

9

Nobody will need to touch it. 5.0 SS in Steam with the rift CV1 doesnt even go that high.

1 Like
10

From the video your colleague posted at the OP :slight_smile:

11

Pretty sure that’s the renderbuffer resolution, before the lens compensation pass. That means it’s not straight up supersampling; it may be chosen to match at wherever the panel resolution as seen through the lens is highest. The effective supersampling factor is variable across the screen because the lens doesn’t show the panel in a rectilinear fashion. Another reason to push the renderbuffer slightly higher is for post-rendering projection adjustment, a component of optimization dubbed things like asynchronous time warp, interleaved reprojection, or brain warp. (It’s not all these things do, but it’s a critical component.)

The hardware reasons xunshu listed are true, and may require tuning across prototype iterations. There may also be software side issues (such as difficulty getting multiple ports to sync up) or protocol issues (such as needlessly large blanking periods - they might be able to cut these for reasons including the dual panel architecture). Protocol tuning brought the SMD ST1080 from 24Hz to 30Hz when running full resolution, which required a firmware update but no driver or hardware changes.

It is also possible that the non-standard (the spec sheet mentions HBR2.5, which I haven’t seen elsewhere, but perhaps DP1.3/HBR3 capable video cards can use it) link rate support for the DP receiver chip could be exploited by a link box. It’s common for link boxes to use a different cabling set, including ones from SMD, Razer and Sony. Pimax hadn’t planned on a link box (it’s now planned as a later add-on), but it would reduce the variables on one side.

I’m glad Pimax are telling us about the status and clearly working on getting it right.

2 Likes
12

o, it is rendering resolutions, different with display resolution and has nothing to do with upscaling.

1 Like
13

@bacon, I believe the question from @evertec still stands. Why do you render it in such a big buffer?

The output res is 25601440, rendering res is 40962657, this gives 1,6 horizontal factor and 1,845 vertical or in other words supersampling factor 1,6*1,845 = 2,95. I assume it is to compensate for pre-lens warp, but this is still quite a lot, more than Oculus or Vive use by default and they are even outputting smaller res to the HMDs.

14

Yes, when I look at the default 1.0 setting on SteamVR for my Oculus it’s 1344x1600 per eye, and the native res of the panel is 1080x1200, so that’s a 1.24 x 1.33 factor. If the input res is 2560x1440, we’re talking a 1.6 x 1.845 factor like you said. That’s putting a strain on the GPU even greater than the native res of the 4K x 2 panels.

1 Like
15

It has to have something to do with keeping detail ie ppd higher in the center. Not only do they need to compensate for distortion, but with 200 degrees of FOV most of the resolution post distortion correction is lost to the periphary. They are probably boosting to compensate for that.

1 Like
16

If they’re losing that much to distortion correction we really will need at least fixed foveated rendering or at least lens matched shading for every game we run on this. That’s 1.741 MP/sec @ the 80Hz they show. I have to set steamvr supersampling at 4.5 to render that many pixels/sec on my oculus. Most games won’t run well at all at that level of supersampling, even on a 1080TI.

2 Likes
17

Lens matched shading was always a good idea. There are other workarounds; for instance, where the pixels per degree escalates one could stencil out a portion of the pixels, and adjust the sampling. This trades fill rate for a bit of extra aliasing. This is the method discussed under “I’m bad at words” in https://alex.vlachos.com/graphics/Alex_Vlachos_Advanced_VR_Rendering_Performance_GDC2016.pdf

1 Like
18

The realization of how many pixels would be needed to maintain detail was one aspect of what initially prompted my swap lens for lower FOV idea in another thread I made.

Those who have tried Pimax 8k and Windows MR have said that the units were about even on actual image quality.

So, yes, its very concerning that they need so much horsepower to.make that FOV truly workable.

Truth be told 120 or 150 degrees would have been a stellar compromise.

1 Like
19

hello @xunhsu just one question, are you using display port 1.2 or 1.4? because only 1.4 can deliver 90 fps, and also are the screens even have the capability to deliver 90 fps on the hardware side…?

1 Like
20

the drawings are per eye right?