Is VR dead?

[gamma123152]

PS5 is mass market console, is VR just a wrong accessory for it at the current price and more importantly lack of content? Chicken and an egg problem, you don't have mass adoption, you can't invest into development.

I have no idea what Sony's strategy is. Maybe they don't intent to sell as many as you think. Adding VR adds A LOT of free marketing for the PS5, marketing that the Xbox doesn't get. Second point is: we don't know what the development of the PSVR2 costed an how much they need to sell before it's profitable, maybe it's not such a big amount since they already have experience with VR. I agree that the pricetag is high for mass market adoption but that doesn't mean that there's no market for it. Maybe Sony is already happy once 5% of the PS5 users buy one, and I think that it's a realistic estimation. Without all this information we cannot say that it's just a wrong accessory imo. And I think that many GT7 players really waited for this, even while it uses motion reprojection; it's a great thing that GT is finally playable in VR. Many Gran Turismo drivers buy a PS5 solely for that title alone.

 

[RCHeliguy]

Today I found out that my 4090 can only drive my Aero in iRacing which is one of the easier to drive Sims at 27ppd. Now it looks pretty damn good, but today I actually tried out 35ppd and holy crap that is SHARP! Unfortunately there is no way any of us will be driving a headset at that resolution without a 50 series GPU ( assuming Nvidia manages another very large jump in performance), Or until the OpenXR foveated rendering with eye tracking becomes a WHOLE LOT more efficient. As for ACC, maybe the 60 series???

Currently native Varjo apps that utilize eye tracking foveated rendering are supposed to handle 35ppd, so it should technically be possible, but realistically unless OpenXR comes to the rescue we've easily maxed out the 4090 without tapping the Aero's full potential.

The reason Varjo has a 39ppd mode is that the lenses are supposed to be high enough quality to see a gain from super sampling. Good luck using that before the 60 series gets here!

 

[gamma123152]

Today I found out that my 4090 can only drive my Aero in iRacing which is one of the easier to drive Sims at 27ppd. Now it looks pretty damn good, but today I actually tried out 35ppd and holy crap that is SHARP! Unfortunately there is no way any of us will be driving a headset at that resolution without a 50 series GPU ( assuming Nvidia manages another very large jump in performance), Or until the OpenXR foveated rendering with eye tracking becomes a WHOLE LOT more efficient. As for ACC, maybe the 60 series???

Currently native Varjo apps that utilize eye tracking foveated rendering are supposed to handle 35ppd, so it should technically be possible, but realistically unless OpenXR comes to the rescue we've easily maxed out the 4090 without tapping the Aero's full potential.

The reason Varjo has a 39ppd mode is that the lenses are supposed to be high enough quality to see a gain from super sampling. Good luck using that before the 60 series gets here!

Sounds like it's going to be impossible to run the Crystal at 35PPD. Since 1. the Aero is more efficient than the Crystal(100% render resolution is lower) 2. iRacing is the most efficient title, Automobilista 2 is more demanding and I want that to run on 35PPD and ACC is probably not even realistic at all. 27PPD what you wrote is what I almost currently run on my G2 with 120% render resolution (but of course it's a bit sharper on the Aero, I expect, since the G2 stops at 2160px per eye, but the difference is probably pretty small with 27PPD at Aero compared to 120% native G2).

These new headsets are the time ahead. If I cannot get the Crystal to run sharper than my G2 then I will probably return the device. Yes the black are better and the sweet spot is better, but it's 1900 euro. This since my main reason for the upgrade is to reduce shimmering further. But the GPUs seem simply not to be capable to do that yet.

 

[Deleted member 197115]

The screen resolution is 2880 x 2720 per eye, everything higher than that is just different degree of supersampling. No different from other headsets, you can kill any card with any headset when doubling native res. And yes, the higher you go, the sharper it looks. I believe around 1.4x was a golden median to restore center resolution back to native after barrel distortion, but going even higher gives benefit of better AA.

 

[RCHeliguy]

This also points to another reason the 12K shouldn't be released yet. All that extra FOV would be for future use when a GPU could actually drive it.

However there is something about the Aero that is a bit interesting in terms of PPD. The average horizontal and vertical ppd is lower than 35ppd. However the ppd in the center of the lenses is actually supposed to be a bit denser than the rest. It doesn't have the extra center screen like the VR-3, but something else is going on.

The actual ppd numbers are interesting if you google for them. What's interesting is that there are conflicting numbers out there.

 

[gamma123152]

The screen resolution is 2880 x 2720 per eye, everything higher than that is just different degree of supersampling. No different from other headsets, you can kill any card with any headset when doubling native res. And yes, the higher you go, the sharper it looks. I believe around 1.4x was a golden median to restore center resolution back to native after barrel distortion, but going even higher gives benefit of better AA.

True, but to achieve 2880x2880px resolution(100% steamVR/OpenXR including barrel distortion) on the Crystal you need to render around 4320x5100px. This while the Aero is something in the late 3000px range to achieve the same (yes small difference vertical). So to render the Crystal with similar PPD will be even more difficult. This while, as @RCHeliguy describes, it's already impossible to render the Aero at full native resolution(or with slight super sampling, what is preferred) with the 4090. So the Crystal is really an HMD that cannot be used at it's full potential yet. Same as the G2, at the release there was also not enough GPU power to render it properly, you need to 4090 to get optimal image quality in sim racing with the G2. History tends to repeat itself... but the extra pixels that the Crystal needs for barrel distortion are the worst in VR history. This makes it way more demanding than the Aero. We will see in the first benchmarks how bad this actually is, because I've also heared that side by side the Crystal is somehow a tad sharper than the Aero(not sure why since it has almost the same resolution so it must have something to do with the lenses).

 

[RCHeliguy]

The Aero is running 4148 x 3556 in 35ppd mode = 14.75 Mp per eye.
The native resolution is 2,880x2720 = 7.83 Mp
In 27 ppd mode it's running 3140x2692 = 8.45Mp about 8% SS

So technically 35ppd is about 75% super sampling.

Your saying the Crystal is likely to be running 4320x5100 = 22Mp because of the amount of cropped resolution.

I'm assuming that is the same as HAM (Hidden Area Mask)

There is an OpenXR setting for foveated rendering where you can turn off rendering the HAM. Ideally that would compensate for the pixels that are not being seen, but I don't know exactly how the performance is affected at this moment.

VR-Compare.com has the following ppd values listed, but as I mentioned I've seen different values.

	Horizontal ppd	Vertical ppd	Peak ppd
Varjo Aero	31.86	33.64	35
Reverb G2	23.58	23.77
Valve Index	14.84	13.19

The peak ppd in the center of the Aero lenses is likely due to how the aspheric lens is shaped such that it is focusing a bit more density towards the center of the viewable area.

The ppd measurements have to do with how much detail you can see. Pixels per degree seems like a simple enough concept. Where physical resolution differs from ppd gets confusing.

I've read that the Aero has lenses that are sharp enough that they hold up to super sampling in a meaningful way where you actually see a sharper image.

We've all seen super sampling on other headsets and know it does look better. "Apparently" the optics on the Aero work particularly well with supersampling. Why, I have no idea. I'm sure the 35ppd peak resolution was measured somehow, but I can't explain it.

 

[RCHeliguy]

I am being told that with the right settings I can in fact drive the Aero at 35ppd. Hmm...

iR Graphics Settings:
MSAA 4x
AF 16x
FXAA/SMAA off (important!)
Sharpening on

Rest of the settings depending on what your CPU can handle. I think with the 13900K you can max out everything you find (except cars on the grid and keep the Sky, grandstands and crowd to low)

---

OpenXR Tool Kit:
FFR,
Outer Ring 100%
165% width
Cull Outer Ring
Inner Ring: 65%
Middle Ring 1/4
Turbo Mode on
CAS sharpening: 70%

---

Nvidia Control Panel:
MFAA on
AA optimizations off
Texture quality: High Quality
Trilinear optimizations: off
AA Transparency: Multisample


Edit: I'm getting about 75 fps with these settings on the track by myself. Still haven't reached 90fps. I'm not getting any flickering and it's smooth, but I'm still not where I want to be. GPU is 100% CPU appears to have breathing room.

 

[RCHeliguy]

OK, I'm obviously still on a learning curve with this.
Some of the settings in iRacing don't take effect until you exit all the way out and come back in again.

I am now getting 35ppd at a solid 90fps in iRacing AND my 4090 is running around 71%.

So there is still a little room left over.... I need to try an AI race now to see how it holds up.

That's rendering 4148 x 3556 and I'm actually seeing closer to 65% and could drop that lower by reducing some settings, but this feels pretty good.

Edit: Tested with a full grid of cars and all is well
The frame rate dropped to mid 80's during a big pile up in a turn, but it was still clean the whole time, no stuttering.

This is a lot sharper than the G2 at 35ppd!

Edit: I would "almost" say there is no point in going over 35ppd. If you are not in a flight sim than there pretty much is no reason to justify going past 35ppd.

In addition, the 13900K & 4090 are driving the Aero well and I don't feel like I need to wait a GPU generation to make full use of it's capabilities.

I'm even mildly optimistic that the Crystal could be driven well with this combination. The 12K might be a GPU generation away from being fully utilized, but given how good 27ppd looks, if the 12K could manage say 160 FOV at 27 ppd, that would still be saying something.

 

[Deleted member 197115]

The Aero is running 4148 x 3556 in 35ppd mode = 14.75 Mp per eye.
The native resolution is 2,880x2720 = 7.83 Mp
In 27 ppd mode it's running 3140x2692 = 8.45Mp about 8% SS

So technically 35ppd is about 75% super sampling.

Your saying the Crystal is likely to be running 4320x5100 = 22Mp because of the amount of cropped resolution.

I'm assuming that is the same as HAM (Hidden Area Mask)

There is an OpenXR setting for foveated rendering where you can turn off rendering the HAM. Ideally that would compensate for the pixels that are not being seen, but I don't know exactly how the performance is affected at this moment.

VR-Compare.com has the following ppd values listed, but as I mentioned I've seen different values.

	Horizontal ppd	Vertical ppd	Peak ppd
Varjo Aero	31.86	33.64	35
Reverb G2	23.58	23.77
Valve Index	14.84	13.19

The peak ppd in the center of the Aero lenses is likely due to how the aspheric lens is shaped such that it is focusing a bit more density towards the center of the viewable area.

The ppd measurements have to do with how much detail you can see. Pixels per degree seems like a simple enough concept. Where physical resolution differs from ppd gets confusing.

I've read that the Aero has lenses that are sharp enough that they hold up to super sampling in a meaningful way where you actually see a sharper image.

We've all seen super sampling on other headsets and know it does look better. "Apparently" the optics on the Aero work particularly well with supersampling. Why, I have no idea. I'm sure the 35ppd peak resolution was measured somehow, but I can't explain it.

188% of native
Native: 2880 x 2720 = 7,833,600
Highest: 4148 x 3566 = 14,791,768
14,791,768 / 7,833,600 = 1.88
That's awfully lot to counter barrel distortion.

 

[RCHeliguy]

I'm going to take a half step backwards on my optimism about driving a Crystal well "today" with a 4090 and top end CPU.

I think iRacing is currently a best case scenario in terms of performance.

How well different titles work is ALL OVER THE MAP!!!

Both my Flight sims (DCS and MSFS) are currently running at 27ppd and still require a slower frame rate 70-89fps. Other people are pushing higher resolutions but living in the 60fps and lower range. The Aero does look good at lower frame rates without stuttering etc, and for people just flying around in MSFS, having a realistic dash and being able to see detail on the ground can be impressive. I saw some of that today.

I can run 35ppd for a very few non-sim games. The big issue with most of the non-sim games is that they do not support OpenXR directly and OpenVR isn't as efficient.

That is "today".

What I'm seeing that is very encouraging is a LOT of effort to find ways to make titles more efficient in VR. There are more titles that will be supporting DLSS 3 in the near future. DCS has announced they are working on support.

OpenXR is also evolving it appears rapidly. There is working DFR ( dynamics foveated rendering ) working for some titles like iRacing, however it only recently started working and optimizations haven't really begun and most titles don't support it.

So I'm optimistic that over the next year we will see improvements in Nvidia driver efficiency, increased support of new features in sim applications and continued improvements in OpenXR.

 

[gamma123152]

There are more titles that will be supporting DLSS 3 in the near future.

I tried all possible variants of DLSS in ACC in VR and also in pancake with Cyberpunk, but it always gives an horrible ghosting effect. This makes it unusable for VR, DLAA comes the closest but it still has to much ghosting. Only if Nvidia gets rid of the ghosting it's worth it, until that it's unusable for VR.

continued improvements in OpenXR.

I think that for the sim racing titles the max. is already achieved with OpenXR toolkit. The developer behind OpenXR toolkit is an perfectionist and he's really driven make it as perfect as possible and I think that he achieved that. So I don't expect much improvements from that seen how good it is already. The only possible thing is that he indeed gets dynamics foveated rendering working in all sim racing titles by a trick, this can still give a huge boost, but from what I understood from him; it's VERY difficult to do because it asks a couple of weeks fulltime work per title to get it done and he's not planning to do that.

 

[metalnwood]

I have the utmost sympathy for your struggles with a 4090 and aero

 

[RCHeliguy]

FWIW my experience with DLSS 3 in MSFS2000 has been good. I think they are using it mostly for detail at higher resolution. I'm not sure if I'm seeing ghosting or not. I'll need to play with it more.

In MSFS and the graphics settings were turned all the way up except the ground detail which was a slider set to 200, but it could go to 400. So I pulled it all the way to 400. Almost instantly I heard my computer fans ramp up to 100%. Keep in mind I have my IEM's in with 37dB of attenuation. I looked around a bit and the visuals were more detailed, but I turned it back down. It just didn't sound healthy.

At 27ppd it's still very sharp. I also tried DCS at 22ppd. Fixed foveated rendering is good for a few fps, but I was actually a bit amazed at how smooth everything looked in DCS even with the frame rate fluctuating all over the place.

There is a wrapper for Dirt Rally 2.0 that is supposed to allow running OpenXR. I haven't set it up yet. At 27ppd it still looks very good and can maintain a solid frame rate.

There is this issue mentioned earlier about blur, ghosting or something having to do with how these displays work. I'm still trying to understand it better. In DCS running mid 70's fps I saw none of this when I looked around. Head motion felt smooth everything looked clear. Sometimes at certain frame rates and certain resolutions it seems more noticable.

One thing I have noticed is that the visuals look sharp and smooth at 22, 27, and 35ppd. I would say that at 22ppd it looks like the G2 does in the center of the screen except there is a larger area in focus.

 

[RCHeliguy]

I have the utmost sympathy for your struggles with a 4090 and aero

This is a bit of cutting edge running on the limits of what current hardware is capable of driving.

I have experienced a few glitches. Yesterday after having my Aero on for a while and bouncing in and out of DCS, switching ppd settings and OpenXR setting and then going into MSFS and doing the same, I saw a USB device not working correctly error and the Varjo's display stopped working. I reinserted the Aero's USB cable and it came right back.

I'm not using SteamVR nearly as much but it still crashes on start occasionally.

The Varjo Base software can be set to support OpenVR and/or OpenXR and I've been using only OpenXR for all the sims except Dirt Rally 2.0 so far. I still need to install another GitHub .dll to allow DR2 to use OpenVR.

I'm currently launching DCS from a batch file using "DCS.exe --force_OpenXR" and have a special wrapper for it the keeps DCS from rendering the 4 displays needed for the VR-3. This is a bug that Eagle Dynamics needs to fix, but someone created a fix for it.

I am very impressed by how well it works in iRacing at 35ppd. I appreciate the efforts of Bernard Berger who worked out the magic settings allowing the Aero to run well at around 65% GPU with a full grid. It does show what this headset is capable of.

While I wouldn't call these "struggles", I am having to jump through a few hoops.

A lot has been worked out already, but a lot more is still being worked on. This video from 10 days ago talks about new "experimental" motion smoothing from Varjo. Notice the small OpenXR frame rate is showing at about 30fps throughout and it still looks smooth.

We have people working on OpenXR wrappers, OpenXR extensions and Varjo is still making improvements and adding features to their base software.

The OpenXR advancements benefit everyone. I don't know if the Varjo smoothing feature is a reprojection like feature or something entirely different.

In some ways things seem mature and in others it still feels like early days.

 

[HorneyMINI]

I'm expecting my PSVR2 shortly and will be my first VR experience. GT7 looks like it's going to be amazing.

 

[GeekyDeaks]

I'm expecting my PSVR2 shortly and will be my first VR experience. GT7 looks like it's going to be amazing.

Mine is supposed to be here today also. I didn't specifically get it for GT7, but it'll be nice to compare it to my PC setup in something like AMS2 as my graphics card alone cost slightly more than the PS5 and PSVR2 combined. I have been deliberately avoiding the hype buildup until I get my hands on it though.

 

[Spinelli]

I'm going to take a half step backwards on my optimism about driving a Crystal well "today" with a 4090 and top end CPU.

The Pimax Crystal is the same resolution as the 8KX but crammed into a smaller area (and a better display) hence the much higher clarity / PPD therefore, from what I understand, it should be around the same GPU-demands wise as the 8K X (at equal refresh rates of course). Lots of people have been running 8K Xs with GPUs like 2080 Tis and 3070s let alone a 3090 let alone a 4090 (but of course a 4080 or higher would only make it better - I'd personally recommend the 3080 as the bare-minimum). The Crystal may even be less demanding than the 8K X due to features in the Crystal not available in the standard 8K X like eye-tracking-based foveated rendering, much less canted displays (meaning much less performance hit in games requiring parallel projections), work being offloaded to the built-in XR2 CPU, foveated transport (not 100% sure if this improves performance), etc.

EDIT: Just wanted to add, because the visuals in the Crystal are so good - the best in the consumer industry (according to just about every one who tried it) - the picture will still look pretty good for those needing to lower the resolution from less than 100%. So, let's assume an 8K X and Crystal run the same native resolution and overall GPU demand, lowering the res to, let's say, 80% will still probably give a good picture on the Crystal whereas doing the same with the 8K X will drop you to a much less tolerable image quality even though both headsets are running the same resolution, GPU-demand, and framerate.

EDIT: This is mostly just personal theoretical speculation, obviously. So take what I say with a grain of salt. When it comes to VR, it's difficult to assume performance because there is lots going on unlike normal monitors which are a particular resolution and that's basically it.

 

[RCHeliguy]

The Pimax Crystal is the same resolution as the 8KX but crammed into a smaller area (and a better display) hence the much higher clarity / PPD therefore, from what I understand, it should be around the same GPU-demands wise as the 8K X (at equal refresh rates of course). Lots of people have been running 8K Xs with GPUs like 2080 Tis and 3070s let alone a 3090 let alone a 4090 (but of course a 4080 or higher would only make it better - I'd personally recommend a 3080 as minimum). The Crystal may even be less demanding than the 8K X due to tech in the Crystal not available in the standard 8K X like eye-tracking-based foveated rendering, having much less canted displays (meaning much less performance hit in games requiring parallel projections), some work being offloaded to the built-in XR2 CPU, tech like foveated transport (not 100% sure if this improves performance), etc.

That sounds great! Fingers crossed.

Are you suggesting that the Crystal will perform better than the Aero? That would suggest that the Pimax base software is more advanced or efficient than the Varjo base software. It is entirely possible given they have been at this a while. Once again fingers crossed!

 

[Spinelli]

Are you suggesting that the Crystal will perform better than the Aero? That would suggest that the Pimax base software is more advanced or efficient than the Varjo base software. It is entirely possible given they have been at this a while. Once again fingers crossed!

That, I have no idea. VR performance is difficult to predict unlike, say, normal monitors which are basically just a resolution.