Apology Letter, project status and risks

Pimax 8K Series
61

Honesty always survive. Thanks for the update. It was clear to me that as organization Pimax needs to be more professional to be able to handle this enormous project. I believe that the technical issues can be solved, although it can take a bit longer then expected, which imo is okay.

I can imaging the situation you are in: a fast growing company, hiring new people, instructing them, developing a challenging product, answering all your backers, being creative all the time, roadshows…when do you sleep?

The reason I was thinking about cancelling my pledge was only for the reason that I had the feeling that Pimax was over his head with this project. But if the organization will be up to the tasks ( and I notice you are working on that) then Pimax will have a major role in the VR world.

As a simracer on PC, VR was the big game changer with the Oculus Rift. Having more FOV, no SDE and higher resolution is very welcome as next step.

Although I do have doubts if my brand new GTX 1080+ will handle the resolution i.c.w. 80-90Hz. With the Oculus and an SS of 1.8 and low/medium game settings (acceptable image) I really need ASW for getting 45 FPS to 90 FPS and would need Brainwarp to do such a trick as well.

My question: do we need as much SS for the 8k as for the Oculus Rift CV1 to have a good image or is it not needed as the rendering to the HMS is already much higher?

1 Like
62

This actually answers my question because 2x4096x2657@80 output for a 1080 is not achievable (?) Hope for a short cut!

1 Like
63

Why would you render at 4k to scale it down to 1440p and then rescale it up to 4k? Doesn’t make any sense.
Is it because of the single DP cable bandwidth limitation?

64

VR tech is very similar to audiophile tech, you should take care about the components and cables materials, shielding and length you are using.
OCC copper, silver and gold are the best friends of the audiophile (and silver solder obviously).
For seated simulation you can provide a 2meters long cable if the perfs are better than 5meters.

2 Likes
65

I’ve been wondering about that myself.

66

I think there is still some room for clarification of issue with HMD V3 Prototype not managing to display 2x4k@90Hz . I think you are throwing DP1.4 word as a mantra, but fact is in specs of the chip ANX7530 used from V3 after on, for 1 DP cable implementation, it clearly states it is not supporting HBR3 which is actually introduced from DP1.3 standard . In specs it is not claiming it can drive 2x2560x1440@90hz which is required for upscaler to properly feed screens at native res and ref.rate. Chip can drive 4k@60 which is not same as 5k@90 required to get to 2x4K@90, in theory max is 2x2.5k@85 but that is not stable as your test show(only 82).

Also problem with forcing the word DP1.4, in FAQ on Kickstarter, GPU required is GTX980/1070, GTX9xx series support only DP1.2, which max out 2*2.5k@~80hz, it can’t achive not even in theory 90Hz at stated input res(2x1440p). So will they have to run @75hz? I read at start of KS that input res can be also 2x2560x1080, that can allow 90Hz?

Demo was done with basically 8k X equivalent because of 2x input cables, so it is not representative of how HMD performs with 1 DP cable(final solution). Problematic is also price dif. from 8k to 8kx, which is 200$(so imposible to at 8K price put 2xDP to solve bandwidth issue of DP1.2 speeds), I think it was fair to say you are demoing 8k X prototypes. Only from V3 we get 8K demos, and they run@75Hz and with limited number of reviews.

Also problem, cable length, I think passive 5m DP will have problem with 2x2.5k@80hz, as they say is stable right now, can you confirm you are testing with 5m cable?

So this is critical post, I know. But I just want to push things forward, that is why I wrote this post. I expect much from Brainwarp, but no demo yet(for simulated 120hz-180hz), and upscaling is interesting for future high res HMDs. So I admit this project is on bleeding edge of tech, but still can be transparent steps on road to great future for Pimax and VR enthusiast as myself!

TL;DR Chip anx7530 is not supporting HBR3, which is part of DP1.3-4 so it can’t support bandwidth for 2x1440p@90hz to upscaler. V1 and V2 were with 2x inputs, basically so far we have 8K X prototypes reviews(200$ price diff). last gen GPU(9xx) support only DP1.2, and can’t run 2x1440p@90Hz,will run @75Hz, or input is reduced to 2x2560x1080@90Hz.

1 Like
67

@Navi_8K The only issue that I can see becoming a real problem for Pimax later on concerning GPU horsepower is that today’s games will run well on a 1080 ti @200 degrees of FOV, but tomorrow’s games may yet be more difficult to render. This is an amazing HMD, but I think Pimax customers will be chasing GPU horsepower like mad for a long time.

1 Like
68

Even the 1080Ti isn’t 1.4 certified, nvidia list their specs as “DisplayPort 1.2 Certified, DisplayPort 1.3/1.4 Ready” (confusingly, the comparison table claims they did get 1.4 in the 1050, but the spec page for that disagrees). The transfer rate target for the Pimax 8K family was always HBR2. What timings are you using to estimate that 90Hz wouldn’t be achievable with DP 1.2 HBR2? Wikipedia lists CVT-R2 2560x1440 60Hz as 5.63 Gbps, so for two displays at 90Hz you’d get 5.63290/60=16.89Gbps, to compare against the 17.28Gbps capacity of quad lane HBR2.

A run with cvt12 produces a plausible but tight mode with ./cvt12 5120 1440 90 -b, using a 714.23MHz pixel clock (limit 720MHz, per pimax post). This does require CVT 1.2 reduced blanking, and frame packing mode (using 180Hz instead of 5120 width) produces an infeasible 768.5MHz pixel clock.

2 Likes
69

Well, it actually is probably achievable with most games, but not at high settings. I’ll have to run some tests on my 1080 with 2.0 supersampling on my Oculus, which would roughly be equivalent to the pixels of the pimax at 2x4096x2657@80. In the meantime, I saw these tests here https://www.reddit.com/r/Vive/comments/601wn8/supersampling_tests_with_a_watercooled_1080ti/. It’s an overclocked 1080ti, but several games still had quite a bit of headroom at 2.0 supersampling. It’s still an insane amount of pixels, and we’d really need next gen cards to be able to run everything at max details unless games make fixed foveated or at least lens matched shading standard to alleviate some of the strain. Having a technology equivalent to ASW would really help too, in some games I can barely notice that it’s running at 45 vs 90, ASW works so well.

3 Likes
70

Thanks, interesting! Tried this with ./cvt12 3840 2160 80 -b
3840x2160 @ 80.000 Hz Reduced Blank (CVT) field rate 80.000 Hz; hsync: 179.440 kHz; pclk: 717.76 MHz
Modeline “3840x2160_80.00_rb” 717.76 3840 3888 3920 4000 2160 2183 2188 2243 +hsync -vsync

So, if they don’t get the 8k stable with more than 80 Hz then they could as well also drop the 1 DP variant and only release a 8k (X) with two DP cables, two scalers and 80Hz (?)

Edit: Sorry, was meant as a reply to the posting of @LoneTech
Edit2: Ok, if they don’t get 2x1440p@90 Hz with 714 MHz stable then 4k@80 Hz with 717 MHz per eye/scaler is probably even more unrealistic then? Sorry, was probably a bad idea…
Edit3: Still unsure how much impact GPU-side upscaling has on GPU requirements (and whether a 2 DP solution would actually benefit from scalers)?
Edit4: This all assuming that the 720 MHz are the upper limit for the scaler, not the panel? (Otherwise the 8k X could also not reach 4k@90 Hz per eye)?

71

I was using wiki page here. You may notice that the “data rate required” (the second table on the page) is not linear to refresh rate. E.g. 25601440@60Hz is not half of 25601440@120Hz. I am not much familiar with DP implementation, so I just assume it may be due to some timing constants which do not scale with refresh rate.

Anyway, using this table I can estimate 2* 25601440@90Hz as 2 (25601440@60Hz + 25601440@120Hz) / 2 = 2* (5,63 Gbps + 11,59 Gbps) / 2 = 17,22 Gbps.
Which is still below theretical limit of HBR2 17,28 Gbps

If however there is some additional latency, or the “constant” does not scale in linear way, you may want to take the value of 2560*1440@120Hz as approximation. Then you get 11,59 Gbps / 120 * 90 * 2 = 17,385 Gbps, which is out of the limit.

We do not know also if gfx card sends two frames of 25601440 or one frame of 51201440 which is then split up and if either operation does not introduce additional delay.

72

Yes, I was also using 1440p@120Hz from wiki as base for calculation…
With better calculation, as @LoneTech did, it shows it can be done in theory, not taking into consideration electrical properties of cable/connections.
They have problem with 85Hz, let alone 90, so this is a sign to me that 90 is just theoretical limit, not usable for cable with length 3m+ . It would be nice if someone from Pimax can confirm if GPU sends 1440p@180Hz to HMD or 5120x1440@90Hz via DP cable?

73

There are indeed a bunch of variables. You know the main three: width, height, and refresh rate. Multiply them to find a minimum required pixel clock. But display hardware must also be told when one image or line ends and the next starts; in CRTs, this was the time the electron beam would swing back. This time is known as retrace, because it goes the opposite way (usually faster), and also as blanking, because you’d keep the image black in that region so you didn’t draw other signals over your image. Further details like where the sync signal triggers within the blanking are defined by the porch times. However, flat panel monitors don’t actually need the retrace, so the timing for it could be relaxed. This led to newer timings including CVT 1.2’s reduced blanking timing version 2, which the Wikipedia page uses (described as CVT-R2). CVT still just defines a common formula for generating modes; knowing the specific limits of your hardware even tighter timing is possible.

So the example modes you’ve estimated from are 2560x1440 at 60Hz or 120Hz. These end up using timings from 2720x1481 and 2720x1525 respectively, both of which have a vertical blanking period of about 0.46ms. The difference thus comes from how many pixels would have fit within that blanking time. One of these bits of overhead is most likely ignored simply by not adding a horizontal blanking between left and right eye images.

A key aspect of this is that the lower you can push the pixel clock, the more margin you gain in the analog domain where signals are actually passing through the wire. Thus improving mode timings by slimming the porches can improve how many real pixels can be passed through the cable. The theoretical minimum pixel clock for the Pimax 5K/8K image is actually 664MHz, far lower than the standard 718MHz. For reference, CVT-R2 standard timings for 82Hz 5120x1440 would be at 648MHz.

Edit: That timing estimate (that lowering pixel clock helps with signal rate) could be wrong for DisplayPort. Per DisplayPort Technical Overview, DP uses fixed transfer rates independent of pixel clock (except the total transfer rate must obviously fit all the bits of the pixels sent). This means even at 75Hz we’re talking about a minimum of 13.27Gbps, and HBR2 (4*4.32=17.28Gbps) must already be in use as HBR is insufficient (8.64Gbps). Thus we’re possibly looking at other limitations such as the supported pixel clocks within the splitter.

3 Likes
74

Personally for me I’m ok with 80-85hz refresh!.

1 Like
75

@bacon @deletedpimaxrep1 @PimaxVR, can anyone tell me if the 5K will have any issues reaching 90hz? From what I understand the limiting factor for the 8K is you are having trouble reaching 90hz in upscale mode to 4k, so with the upscaler having been eliminated on the 5k can we expect a full 90hz refresh rate?

Also, will the 10m cable ship with the HMD? I’m only asking because we’re being asked to pay $10 extra for shipping a cable and that’s the only way it makes sense to me.

76

Thank you, ,really appreciate. ( hard for the long time for the wireless :frowning: ) but thanks for your transparency

77

Sorry if you didn’t want my answer, but I read that 5K has less problems getting to run @85Hz, so far. Someone from pimax team wrote, lazy to quote now :slight_smile: . I have a theory that 90Hz will be possible only by using 2xDP, because even tho they write DP1.4, it is using chip that don’t support HBR3, so it runs at DP1.2 speeds, and at 2x1440p@85 we are very close to theoretical limits of DP1.2, for scaler maybe to close(8K so far stable at 82Hz)…

10m DP cable passive is not possible I think. 5m is included but not sure if they even done testing with 5m DP cable when doing this risks analysis. For 10m will be required breakout box - sold after KS (my assumption).

1 Like
78

The thing with the wireless is if it takes a year and then costs 100 usd more than a tpcast i will print out my 100 dollar voucher and eat it in rage.

79

@deletedpimaxrep1 , thanks for the risk analysis, as I am a 8K X backer, could you please publish similar risk analysis report for 8K X as well? Appreciated!!! Thank you.

80

I’m not going to spend an hour reading through you and other peoples speculation when I can simply ask the people who are actually making the product; so in that sense you’re right I didn’t want your answer. In a previous post (see here for more info) @PimaxVR mentioned the issue was that “The Driver IC’s capability upper limit is ~90Hz under upscale mode”.

2 Likes