Help with performance with Crystal + 4090

[WipeUout]

5 hours ago, mbucchia said:

Look up "vr barrel distortion" on google. Every device does this, you render at a resolution higher than the physical display (typically 1.4x) to account for the pixel distortion created by the optics. 

I don't understand why the PimaxXR reported resolution is lower than my native?  I am at "1" everywhere (Pimax client =1,  OXRTK no override, DCS PD=1)

I have a PimaxXR reported resolution of 5012x3160 (15.8 MP) which is a bit lower than the physical resolution of the 8kX 4k panels x 2 at 7680x2160 (16.6 MP).   Unless the reported resolution is per eye?  This would mean that the upscaled "for barrel distortion correction" is 190%? 

[mbucchia]

8 hours ago, WipeUout said:

I don't understand why the PimaxXR reported resolution is lower than my native?  I am at "1" everywhere (Pimax client =1,  OXRTK no override, DCS PD=1)

I have a PimaxXR reported resolution of 5012x3160 (15.8 MP) which is a bit lower than the physical resolution of the 8kX 4k panels x 2 at 7680x2160 (16.6 MP).   Unless the reported resolution is per eye?  This would mean that the upscaled "for barrel distortion correction" is 190%? 

Yes the value in PimaxXR is per eye.

[WipeUout]

45 minutes ago, mbucchia said:

Yes the value in PimaxXR is per eye.

After second thought, it makes more sense.  I am using wide FOV (160 degrees H) and it would take a much more than 30% bigger image to match the projected image through the lens.  Good thing that this operation to apply barrel distortion correction is done by shaders, rendering an image at 190% would seriously reduce the FPS to less than 25% of what it is in 2D. 

[dlder]

On 8/4/2023 at 8:45 PM, WipeUout said:

Getting the same problem with my 8kX

Thanks for confirming that this is a "normal" issue^^

On 8/4/2023 at 8:45 PM, WipeUout said:

Pimax FFR does not work with the OpenXR runtime, it will only run with SteamVR.

Thanks, good to know!

On 8/4/2023 at 8:45 PM, WipeUout said:

Are you super sampling some where?

Nope; these numbers - Render Target / Pixel Density - do not correlate with the resolution of the displays. I still have to find out, how:

• Render Target = 1 = 4.312 x 5.102

correlate with:

• Display Resolution = 5.760 x 2.880

I've tried a Render Target of 0.75, which would only be 12,37 MP -> the FPS were of course fantastic. But the picture quality isn't what I like about this HMD: "Crystal" clear picture^^

-> If you have a lower one, then you are probably using "balanced" in the Pimax-Client too... not "maximum", which would be Render Target of "1"

 

 

21 hours ago, Nedum said:

I myself still waiting for my crystal, but I'd read that many guys run with a resolution around 3300 x 3800 (done in the OpenXR Toolkit, set size to 75%(Performance Tab))

Well, of course one can lower the number of calculated Pixel; I just don't see the point. Why buy a high resolution HMD if you are not rendering 100% of those pixels; UpSampling like NIS (which OpenXR can do) is possible and not as bad as just lowering the resolution, but it's still not DLSS, which imho might be up to par quality wise to 100% resolution.

 

17 hours ago, mbucchia said:

Look up "vr barrel distortion" on google. Every device does this, you render at a resolution higher than the physical display (typically 1.4x) to account for the pixel distortion created by the optics. 

AWESOME, thank you!
I hope that will explain what's happening
I just now from testing that lower then 100% isn't an option. It's like in 2D, but even more apparent, because the displays are right infront of your eyes. So 100% resolution + 2xMSAA is the least one can use.

 

16 hours ago, 72Stu said:

tracked foveated rendering

One can only hope, but testing with FFR makes me believe differently. You just can't set the quality so low that you'll gain like 20% FPS or the visuals will be greatly degraded!
-> so I doubt it will be a game changer. Yes, everyone hypes about the newest and greatest tech / feature, and Hope Dies Last, but I'm not too optimistic.

 

Good thing I bought the 4090 which is way, way too expansive (still), but with these many pixels to be rendered, I doubt that any type of optimization will substantially improve performance.
I've tried a lot.
I will try more (from the videos that were suggested).

I'm now playing with "High" graphics settings and have 40 (ground) to 90fps 4000ft and up (on Caucasus though^^)

 

Custom2.lua

[Nedum]

4 hours ago, dlder said:

Well, of course one can lower the number of calculated Pixel; I just don't see the point. Why buy a high resolution HMD if you are not rendering 100% of those pixels; UpSampling like NIS (which OpenXR can do) is possible and not as bad as just lowering the resolution, but it's still not DLSS, which imho might be up to par quality wise to 100% resolution.

DLSS will (for a very long time) not be an option for people who like a crystal clear picture. If you own the MSFS2020 you can test it by yourself. It's not an option, not a tiny little bit.

There is no GPU on the market, which can give you the performance you need to fire up a Pimax Crystal in DCS with the best visuals. You have to tweak it. And you have to look for your personal best visual settings to run DCS smooth.

You have to find a way to optimize your settings to stay lower than a 20 ms frame time with motion smoothing or not higher than 10 ms without motion smoothing to get a smooth game experience.

Is that your first experience with DCS and VR?

 

[dlder]

17 hours ago, Nedum said:

Is that your first experience with DCS and VR?

That it is; I'm still using the mouse to adjust controls, as the controllers don't work very well. That's probably partly because controllers are a bit weird and cumbersome in DCS anyway, and partly because the controllers and the tracking is pretty garbage (I've only used the original Vive 1 wands).

[dlder]

On 8/5/2023 at 5:54 PM, mbucchia said:

Yes the value in PimaxXR is per eye.

Wait... what? But that would double the amount of calculated pixels?! That sounds... excessive.

4k-UHD is "3840 x 2160", meaning a meager 8.29 MP.

PimaxXR says 4.312 x 5.102. If this is the value per eye, that would mean it's not 21.99 MP (seems with 20.98MP I've miscalculated the last time), but double that? So 44 MP? That would be more then 8K-UHD (7.680x4.320 = 33.18 MP)... 

[nikoel]

1 hour ago, dlder said:

Wait... what? But that would double the amount of calculated pixels?! That sounds... excessive.

4k-UHD is "3840 x 2160", meaning a meager 8.29 MP.

PimaxXR says 4.312 x 5.102. If this is the value per eye, that would mean it's not 21.99 MP (seems with 20.98MP I've miscalculated the last time), but double that? So 44 MP? That would be more then 8K-UHD (7.680x4.320 = 33.18 MP)... 

So unfortunately Pimax makes it very difficult to understand resolutions. I suspect this is by design, they are after-all lying about having 8K resolution. One way to muddy the waters is just to throw enough numbers around

First lets start with the resolution of the screens. There are two displays that are 2880 x 2880 per eye - aka 4K resolution. Some outlets and pimax in their software report this resolution as combined across the entire view spectrum. Which becomes 5760 wide and 2880 high. This is where the confusion comes from. Their 8K is actually a sliver over 5K

From there on you can apply your super-sampling rates. In the past with fresnel lenses 1.3-1.4x was the point of diminishing results. Take Reverb G2 with 2160 x 2160 displays, which at 100% Resolution had the 1.4x multiplier pre-applied giving us around ~3100x3100px (cant recall the fully exact number) Unfortunately my friend returned the headset before I could figure out if the same multiplication ratios applied to the crystal - not that it mattered as the performance at those resolutions was not good. He "only" had a 3090ti

Edited August 7, 2023 by nikoel

[WipeUout]

2 hours ago, nikoel said:

From there on you can apply your super-sampling rates. In the past with fresnel lenses 1.3-1.4x was the point of diminishing results. Take Reverb G2 with 2160 x 2160 displays, which at 100% Resolution had the 1.4x multiplier pre-applied giving us around ~3100x3100px

My understanding is that it is not super-sampling but barrel distortion correction which is specific to each lenses and HMD combination.  Super-sampling means rendering at a higher resolution and this is not the case with barrel distortion correction as the HMD stretches the image by using shaders, which is much less demanding from the GPU.  My 8kX has a factor of 1.9 (190%) when I use the full 160 degree wide FOV.  Good thing shaders do that job, else my FPS in VR would be less that 1/4 of what it is in 2D.

As for the Pimax way to advertise resolution, not very honest indeed... nevertheless, I enjoy my 8kX, it is simply stunning to sim with this wide FOV. 

 

3 hours ago, dlder said:

PimaxXR says 4.312 x 5.102. If this is the value per eye, that would mean it's not 21.99 MP (seems with 20.98MP I've miscalculated the last time), but double that? So 44 MP? That would be more then 8K-UHD (7.680x4.320 = 33.18 MP)... 

If PimaxXR reports that resolution, it could mean that you are Supersampling a lot your image.  I doubt very much that the Crystal has a barrel distortion correction ratio of 21.99/8.3= 265%.  You are overriding the resolution in your game/render quality setting in Pimax client, probably set to "maximum" or "customize" with a value higher than 1.

Edited August 7, 2023 by WipeUout
Typo

[dlder]

31 minutes ago, WipeUout said:

My understanding is that it is not super-sampling but barrel distortion correction which is specific to each lenses and HMD combination.  Super-sampling means rendering at a higher resolution

That's what I always thought to be true. Super Sampling in VR means a Pixel Density of over 1 (= 100%). But as I don't do that and I still have an extremely high resolution, this really has to be the barrel distortion.

32 minutes ago, WipeUout said:

this is not the case with barrel distortion correction as the HMD stretches the image by using shaders, which is much less demanding from the GPU.

Really? I haven't yet had the time to research barrel distortion, but that sounds... unlikely. The only thing that nVidia was/is working on was called "MVR" - Multi-View Rendering, but no game every utilized that feature.
https://developer.nvidia.com/vrworks/graphics/multiview

47 minutes ago, WipeUout said:

a value higher than 1.

Nope, nowhere is it set beyond 1 or 100% pixel density.
That is exactly my point: where is this resolution coming from. I'm gonna try my luck on the official Pimax Discord.

 

Again, thank you all for your valued input!
It's quite interesting to see how this all is working (this is not just for DCS; it's VR in general).

[nikoel]

19 hours ago, WipeUout said:

My understanding is that it is not super-sampling but barrel distortion correction which is specific to each lenses and HMD combination.  Super-sampling means rendering at a higher resolution and this is not the case with barrel distortion correction as the HMD stretches the image by using shaders, which is much less demanding from the GPU.  My 8kX has a factor of 1.9 (190%) when I use the full 160 degree wide FOV.  Good thing shaders do that job, else my FPS in VR would be less that 1/4 of what it is in 2D.

As for the Pimax way to advertise resolution, not very honest indeed... nevertheless, I enjoy my 8kX, it is simply stunning to sim with this wide FOV. 

 

If PimaxXR reports that resolution, it could mean that you are Supersampling a lot your image.  I doubt very much that the Crystal has a barrel distortion correction ratio of 21.99/8.3= 265%.  You are overriding the resolution in your game/render quality setting in Pimax client, probably set to "maximum" or "customize" with a value higher than 1.

 

Yeah I agree with your assessment of super-sampling. It's annoying that there is no true industry standard when it comes to reporting of resolution. For instance with Reverb G1 100% resolution in SteamVR is 2160 x 2160px even though it uses exactly the same LCD panels as the G2 - which reports it's 100% resolution as 3100 x 3100px (1.4x corrected). My Quest Pro has two screens with resolution of 1800 x 1920px - Oculus software combines the two screens but gives me 1x resolution of 4128 x 2096

I have always found it best to figure out what is the "native" resolution of the screen(s) and then calculate the barrel distortion from there and then do whatever you need to do to get there, preferably in one place. Say the 1.4x resolution to overcome the distortion and use that as the general guide for what to aim for with a caveat that the pancake lenses and different optics will give different results

In the end those are just numbers to keep in mind, knowing the inner workings of how resolution works within the VR space avoids these pitfalls

Edited August 8, 2023 by nikoel

[WipeUout]

6 minutes ago, nikoel said:

In the end those are just numbers to keep in mind, knowing the inner workings of how resolution works within the VR space avoids these pitfalls

Yes, I agree.  Barrel distortion correction can be very confusing and should not be the focus as it is not a variable in the user-controlled performance equation.  Rendering resolution relative to the physical HMD resolution is the key element to understand and where the very bulk of the GPU's work is done.  

[dlder]

I totally agree!

This has been a more of an academic question in the end, because you can't do anything about it, other then lowering the pixel density (not really viable) or upsampling (which is better, but still not that great).

I just was baffled by, why the 4090 wouldn't perform as well as expected; well, I didn't expect such a large amount of pixels to be rendered tbh 

And although playing in VR is very cool indeed, I'm still not sure I'll switch to it.
1) the graphics in 2D are still better
-> there isn't any pixelated shimmer from medium to long distance. It would need a way higher resolution to combat that... so, 6090 maybe?

2) using the mouse and not your hands/fingers to actuate buttons is viable and better then using the controllers, but using physical buttons (UFC, MFDs) is just way better
-> as long as you cannot reliably use 2 haptic feedback gloves (and from what I hear, this works still wonky at best) or you can use the cameras to bring your physical controls into DCS (which would pose another problem, because nobody has a 1:1 replica of all the planes at home), physical buttons are still best

[nikoel]

1 hour ago, dlder said:

I totally agree!

This has been a more of an academic question in the end, because you can't do anything about it, other then lowering the pixel density (not really viable) or upsampling (which is better, but still not that great).

I just was baffled by, why the 4090 wouldn't perform as well as expected; well, I didn't expect such a large amount of pixels to be rendered tbh 

And although playing in VR is very cool indeed, I'm still not sure I'll switch to it.
1) the graphics in 2D are still better
-> there isn't any pixelated shimmer from medium to long distance. It would need a way higher resolution to combat that... so, 6090 maybe?

2) using the mouse and not your hands/fingers to actuate buttons is viable and better then using the controllers, but using physical buttons (UFC, MFDs) is just way better
-> as long as you cannot reliably use 2 haptic feedback gloves (and from what I hear, this works still wonky at best) or you can use the cameras to bring your physical controls into DCS (which would pose another problem, because nobody has a 1:1 replica of all the planes at home), physical buttons are still best

Eh, you get used to the shimmer. Once you realise it's the way of the lands, its quite okay actually - also the HMD has a lot to do with it too. For instance the Quest Pro has it's screens counter-rotated by around 21` - but yes, the pixel density of even a 1080 screen is far higher than any HMD as it presents across a far smaller real word view spectrum

PointCTRL is what you're looking for. Unfortunately the guy who manufacturers them does it more as a side-hustle than say a proper business and the waiting line is rather long. At least it gives you a perspective of what it was like waiting for a car in Soviet Union. With that said, I have one and it's far, far superior to mouse. I also have a full WinWing suite including the screens and can navigate them all including the MFD buttons all by touch. (It also helps that I can also blink down through the small opening below my vision line to see the physical buttons if I need them) 

I am not currently playing DCS as ED have butchered my two favourite planes into unusable realm, but according to the bug reports they should have them fixed on the next update. So I am looking forward to jumping back in again

Edited August 10, 2023 by nikoel

[dlder]

PointCTRL would be an option, but alas, no haptic feedback; it would be fantastic if you could actually "feel" the buttons and switches you press

And yeah, the wait time... Unreal(tm)^^

[WipeUout]

1 hour ago, dlder said:

And although playing in VR is very cool indeed, I'm still not sure I'll switch to it.
1) the graphics in 2D are still better
-> there isn't any pixelated shimmer from medium to long distance. It would need a way higher resolution to combat that... so, 6090 maybe?

3D will look as good as 2D the day we have:

-HMD that match or surpases the humain eye equivalent PPD which I read somewhere is around 60-70 PPD.

-Hardware that can render the huge amount of pixels required to get 60-70 PPD at a decent level of details and FPS rate.

Until then, 3D will always look less detailed than 2D.  Nevertheless, the VR proposition today is to trade quality of visuals for a higher level immersion and much more realistic experience.

[dlder]

18 hours ago, WipeUout said:

the VR proposition today is to trade quality of visuals for a higher level immersion and much more realistic experience.

This. 100% ack!

That's why "made for VR" games can be soooo good (especially of course FPS).