Performance improvement for all VR Headsets: Quad View Foveated rendering

Emir_de_Passy

Active member
Please consider implementing Quad View Foveated rendering in LMU as it would be beneficial for ALL VR HEADSETS. (Not just eye tracked or expensive ones).

For those who are unfamiliar, it is a form of Foveated rendering that involves two small high-resolution focus areas and two low-resolution peripheral areas, with the focus areas in the center, where the gamer is looking directly.
It is the same principle as the foveated rendering we have been using with the OXR toolkit or VRperfkit, except it reduces pixel count while the variable shading rate (VRS) method used in OXR toolkit and VRperfkit adjusts the level of detail within a pixel block without changing the overall pixel count, which is why QVFR is significantly more powerful. People sometimes believe that it's exclusive to eye-tracked headsets, but it's beneficial for all types. The only difference is that, with an eye-tracked headset, you can make the focus areas smaller. QVFR is more CPU intensive (4 views to render instead of 2) but saves a lot of GPU workload. For example, Pimax could go from 5 to 55fps using DCS with a Crystal and an RTX 2060. According to the modder who implemented it in Richard Burns Rally, it was easy,
Here is a link that explains more; hopefully, you can determine if it is easy to implement in LMU.
https://github.com/mbucchia/Quad-Views-Foveated/wiki


- Foveated rendering is great for performance
- VRS is easy and cheap on the CPU
- QVFR involves doing a second render pass for the foveated part at a higher resolution then merging, which is a cpu cost, but greatly reduces GPU load
- both the above techniques work with both Fixed Foveated rendering and eye-tracked foveated rendering.
- On the headsets that support it, eye-tracked foveated rendering is a great way to go

A second render pass for the foveated part' is the same high-level thing that's done with XR_VARJO_quad_views but in no way requires it. If this is done in the engine, it works everywhere, it can be always on, and because it works everywhere and is always on, doesn't increase the amount of different configurations that needs to be tested.



If you want to do the composition yourself, here is a link: https://registry.khronos.org/OpenXR/specs/1.0/html/xrspec.html#XR_EXT_eye_gaze_interaction
This also makes it work on every headset without an API layer, but is more involved than what the RBR dev did.

Quad view foveated rendering is also available as an extension in OpenXR 1.1 (under the name Stereo with Foveated Rendering) https://www.khronos.org/news/press/...ther-streamline-cross-platform-xr-development


Another way to implement would be to support quad views in an API layer as the RBR modder did.

This would be easier than doing the composition yourself but this would only make it work with the list of headsets below (already quite extensive). This also requires an API layer whereas the first option doesn't.


Quad view foveated rendering in LMU would make it the de facto best VR sim racing implementation by a significant margin. It would guarantee to attract new VR users, many of whom use other titles mostly because they run well in VR. (e.g. AMS2, Iracing). QVFR would make LMU easier to run in VR than on triple screens and even on Ultrawide screens and, therefore, accessible to people with more modest PC configurations.

Here is a link to a discussion about various possible implementations:

www.reddit.com/r/Pimax/comments/1gisyv8/comment/lvdy3o2/

P.S. Apologies if I wrote any inaccuracies, as I am not a specialist.

UPDATE APRIL 2025: Microsoft Flight Simulator 2024 now supports QVFR and is the first current sim to do so not by accident. (Richard Burns Rally supports QVFR but is done by a modder)
Their implementation works on every headset without the need for any mod or special headset software, just like I wrote it could be done above. The GPU performance gain is about 30% and hopefully will improve when out of beta status. This is done at the cost of CPU performance, as expected.
Here is a technical summary of how they did it, from Mbucchia, the QVFR God.


UPDATE MAY 2025: IRACING now supports QVFR with their Season 3 2025 update. And is the second current sim to do so not by accident. The method used is a 2018 Nvidia API called Multi View Rendering which expands on Single Pass Stereo, which was already implemented. It's extremely effective and has a small CPU cost. However it only works with nvidia GPU.

 
Last edited:
Did the LMU developers EVER reply on this most important VR subject? On what the reason is that they haven't implemented this yet?

According to the "inventor" of this software technology, Matthieu Bucchianeri; it's less than a week work to implement this for a good developer.

So is it a lack of skills or time at Studio 397 that this isn't implemented yet?? Or are there engine limitations that create issues to implement quad views? Or does Studio 397 not understand the importance of this? I cannot believe the last, so it must be one of the first three.

I ask this without the intention to disrespect the dev's. I am honestly curious because quad views should be a "no brainer" to implement in this game. It's by far the most easy way to boost the fps in VR massively without an complete rewrite of the code and there are to my knowledge no real downsides.
 
Did the LMU developers EVER reply on this most important VR subject? On what the reason is that they haven't implemented this yet?

According to the "inventor" of this software technology, Matthieu Bucchianeri; it's less than a week work to implement this for a good developer.

So is it a lack of skills or time at Studio 397 that this isn't implemented yet?? Or are there engine limitations that create issues to implement quad views? Or does Studio 397 not understand the importance of this? I cannot believe the last, so it must be one of the first three.

I ask this without the intention to disrespect the dev's. I am honestly curious because quad views should be a "no brainer" to implement in this game. It's by far the most easy way to boost the fps in VR massively without an complete rewrite of the code and there are to my knowledge no real downsides.
I'm not even sure the devs are aware of this request. Even if they aren't, surely Pimax should make them aware of the option.
Regarding your hypotheses, I doubt it's a lack of skills because as I was told by several openxr devs, it shouldn't be too hard to implement.
Time, or rather the willingness to dedicate dev time to the VR mode is probably the issue.
To be fair, given the delay with the dlc, driver swaps implemention etc, I can understand that they need to focus in other places than VR at the moment.
One thing I'd love studio 397 to change is the mindset of "how can we serve the VR community that represents 5% of our users" to "how can we grow the VR user base from 5 to 15-20%?"
Iracing quad view implemention has done more than we could ever accomplish with this feature request. I've read so many users stating that they won't need to upgrade their GPU thanks to the gains from QVFR. Instead, they can now buy a newer headset. VR with QVFR and a headset like a pimax crystal light is now cheaper and in some ways more immersive than a triple screen set up. And the funny thing with Iracing is that it's fixed and it is nearly unnoticeable. I could disable eye tracking in my headset forever if all QVFR implementions are like theirs. It turned an ugly duckling of a game into a decent looking one. Quite impressive.
 
Interesting thoughts.

"I can understand that they need to focus in other places than VR at the moment".

Why should they focus on secondary things first? I see zero reasons for that. Quad Views Foveated Rendering/optimal VR performance should be a basic thing for any self respecting sim title. This sim is on the market for more then a year now.
If it's indeed true that adding Quad Views costs them less then a week to implement; it should have been done first before any DLC development or driver swap implementation; that are secondary/way less important things to have in this sim.

Why do the dev's not comment on this if it's this easy or not to implement?

I understood from Jaap of Pimax that he also hoped that QVFR gets implemented in this sim now that the investment is done.

I wonder what S397 answered about this once Pimax brought up the subject while transferring the 2.5 million to their company. This subject must have been spoken about.
 
Hello dear VR & QVFR fans & interested parties 🖐️ ,

personally, I think it would be counterproductive if a "real" developer were to comment on this,
because then in the future, people would always insist that someone had already commented on QVFR,
even with other suggestions.

Perhaps :unsure: I'm misjudging this & developers have already commented on other things here,
but I haven't seen any responses from real developers in the posts I've read so far,
which is certainly only a small portion of all posts.

So for us, all we can do is wait & see and keep our fingers crossed.

However, I believe that the VR sector will definitely expand,
because thanks to SimRacing & SimFlight, VR is increasingly moving
from a niche to the mainstream & that's a good thing in this case (y)

Happy Sunday everyone✌️
 
This is not very useful if you have good lenses and no eye tracking like Quest 3. Also in DCS, using this heavily impacts CPU frametimes and overall it's better not to use it but my impression is that DCS's implementation has this weakness and it could be done better.

Just want to point out that indeed as you suspect this is not an issue that is inherent with quad views, it is only inherent to a poor or at least less than optimal implementation.

The reason for the CPU overhead is due to a lack of instancing, which then requires the CPU to draw the geometry multiple times. This already results in more overhead in normal VR (any non instanced geometry will have to be drawn twice), and is exaggerated in quad views as any shared geometry between the views will have to be drawn 4 times.

Devs taking the time to instance the geometry removes this overhead, and as demonstrated by the recent iRacing implementation (where they have instanced 100%) there is no additional CPU overhead at all.

While iRacing has implemented this using the somewhat archaic proprietary Nvidia MVR, I would guess this is likely a stop gap as once they move to DX12 all the things that can be achieved in DX11 + MVR (and more) can be achieved natively.

It does still stand as a rather interesting case study - even in its current fixed format (which obviously brings no additional benefit to eye tracked headsets vs others) the gains are extremely significant and difficult to notice most of the time at default settings which can also be further adjusted to user preference.

So I would suggest that iRacing does in fact stand as a useful demonstration of exactly what emir was talking about - a universally beneficial implementation of QVFR for all VR headsets. No additional CPU overhead, significant performance improvements for GPU side even when used in fixed mode, and adjustable parameters in case you happen to need to increase the size/resolution of the foveated areas on a given headset. Obviously the larger you make the foveated area/the higher the peripheral resolution the lower the improvements will be, but as there is no penalty on CPU any gain ultimately is like a free GPU upgrade even if you are bringing it down to 20-30% rather than 60-70% or more.

It will likely only get better as they move to DX12 and away from Nvidia’s MVR, and once eye tracking is added (whether that be through iteration in the current version or delayed until the new engine) then it will become even more performant/hard to notice for those that have suitable headsets.

I am really glad that iRacing has shown the way with this, as even if one isn’t a fan it does act as a wonderful case study when trying to convince of the potential benefits when a dev implements it on purpose, and with all their ducks in a row.
 
Last edited:
Just want to point out that indeed as you suspect this is not an issue that is inherent with quad views, it is only inherent to a poor or at least less than optimal implementation.

The reason for the CPU overhead is due to a lack of instancing, which then requires the CPU to draw the geometry multiple times. This already results in more overhead in normal VR (any non instanced geometry will have to be drawn twice), and is exaggerated in quad views as any shared geometry between the views will have to be drawn 4 times.

Devs taking the time to instance the geometry removes this overhead, and as demonstrated by the recent iRacing implementation (where they have instanced 100%) there is no additional CPU overhead at all.

While iRacing has implemented this using the somewhat archaic proprietary Nvidia MVR, I would guess this is likely a stop gap as once they move to DX12 all the things that can be achieved in DX11 + MVR (and more) can be achieved natively.

It does still stand as a rather interesting case study - even in its current fixed format (which obviously brings no additional benefit to eye tracked headsets vs others) the gains are extremely significant and difficult to notice most of the time at default settings which can also be further adjusted to user preference.

So I would suggest that iRacing does in fact stand as a useful demonstration of exactly what emir was talking about - a universally beneficial implementation of QVFR for all VR headsets. No additional CPU overhead, significant performance improvements for GPU side even when used in fixed mode, and adjustable parameters in case you happen to need to increase the size/resolution of the foveated areas on a given headset. Obviously the larger you make the foveated area/the higher the peripheral resolution the lower the improvements will be, but as there is no penalty on CPU any gain ultimately is like a free GPU upgrade even if you are bringing it down to 20-30% rather than 60-70% or more.

It will likely only get better as they move to DX12 and away from Nvidia’s MVR, and once eye tracking is added (whether that be through iteration in the current version or delayed until the new engine) then it will become even more performant/hard to notice for those that have suitable headsets.

I am really glad that iRacing has shown the way with this, as even if one isn’t a fan it does act as a wonderful case study when trying to convince of the potential benefits when a dev implements it on purpose, and with all their ducks in a row.
I couldn't have written it better.
 
What's the latest on this?
I've heard crickets.

Except Pimax has asked on various discords/reddit what features would sim racers like to see implemented.

They also added in parentheses that they may or not be in contact "with some sims to improve VR". Given their ties to LMU, take this how you wish.
With this feature request, but most importantly msfs2024 and Iracing implementing QVFR with success and of course Pimax being a partner, it's very likely s397 knows about QVFR and its benefits. Now that the driver swap hurdle and 1.0 status have been cleared, let's hope they dedicate some time to VR.
 
Really interesting to think about how this is beneficial to non eye tracking headsets.. I mean on my pico 4 anything outside the sweet spot isn't really worth looking at. You need to move your head anyway.

Having a hot sweaty head and a slightly smeary view on vr makes me feel like a real racing driver with a helmet on 😂
 
Really interesting to think about how this is beneficial to non eye tracking headsets.. I mean on my pico 4 anything outside the sweet spot isn't really worth looking at. You need to move your head anyway.

Having a hot sweaty head and a slightly smeary view on vr makes me feel like a real racing driver with a helmet on 😂
Iracing supports quad view foveated rendering right now and it's not eye tracked. Unless I look for it, I never notice it.
Next month, they will enable eye tracking but it's already outstanding as it is.
Sane for msfs2024. Initially it was fixed and worked great for everyone.
That's the only way to do VR. Any other way is inefficient as there is no point rendering the whole image at 100% resolution
 
Hi all, yes we need foveated rendering of course +1.
Thanks for this post & hope devs & Pimax support will help to bring this real.

Currently, I'm forced to crop the render resolution to enjoy LMU. I'm using only maybe 60% of my headset's FOV... it's enough to enjoy but defintely can be improved to reach a better immersion.

And by the way, all this is possible using mods ... not natively, glad there are talented people out there that bring us those tools ... (Oculus Tray Tool, OpenComposite, OpenXR toolkit...)

Luckily, LMU @72Hz is running fine but I know from other sims that racing @90Hz or higher is mind blowing and put less effort on eyes. For example, I cannot race for more than 1 hour @72Hz currently, then my eyes are tired.
 
Hi all, yes we need foveated rendering of course +1.
Thanks for this post & hope devs & Pimax support will help to bring this real.

Currently, I'm forced to crop the render resolution to enjoy LMU. I'm using only maybe 60% of my headset's FOV... it's enough to enjoy but defintely can be improved to reach a better immersion.

And by the way, all this is possible using mods ... not natively, glad there are talented people out there that bring us those tools ... (Oculus Tray Tool, OpenComposite, OpenXR toolkit...)

Luckily, LMU @72Hz is running fine but I know from other sims that racing @90Hz or higher is mind blowing and put less effort on eyes. For example, I cannot race for more than 1 hour @72Hz currently, then my eyes are tired.
Thank you for the support. Just to clarify, what's possible by using mods is Variable rate shading foveated rendering. Quad view (significantly more efficient than VRS) unfortunately requires developer involvement. And the earlier it is implemented in the pipeline, the more effective. E.g. Iracing.
 
Yes, foveated rendering via openxr tool etc only benefit around %5-10 perfrmance gain.
I have psvr2 and donot use eye tracking by using pimaxmagic4all, performance gain is eligable,
by using modded openxr toolkit and crop2fov feature, an aggrassive up and down reduction of image, i can use 120hz and mostly ultra settings with 8x msaa.
 
Hi everyone,
I’ve just tested mbucchia/Quad-Views-Foveated and the latest version of the PSVR2 toolkit, but unfortunately in my case I didn’t see any improvement – in fact, performance was noticeably slower.
Given that LMU is financed by a VR company, it’s even more surprising that this basic functionality hasn’t been implemented yet, especially considering that iRacing already has it. Right now, many players are forced to crop their view, and surely many can’t even play in VR at all. Adding this feature could enable tens of thousands more players to enjoy LMU in VR, while also improving the experience for existing VR users.
Rendering the whole view at maximum resolution is insanely inefficient. And OpenXR already provides the necessary stereo layers / projection layers to implement foveated rendering and multi-view optimizations.

Does anyone know if the developers are already working on this or if it’s on their roadmap? It would be reassuring to know whether this request is being considered, since the benefits are so clear.
 
100% agree that this is a must have improvement to the game for VR users. I have a very high end system and still struggle to keep 90hz on my crystal super and with other implementations of quad views/DFR such as iRacing it makes a huge difference. I know Pimax backs this project and should be pushing for DFR to be implemented.
 
Knowledgable VR folks: Do you know what the effect is (just inside the headset, not capture) when the edges of the geometry will "wobble"? What is this evidence of? Is my Valve Index just struggling?
 
Yes, foveated rendering via openxr tool etc only benefit around %5-10 perfrmance gain.
I have psvr2 and donot use eye tracking by using pimaxmagic4all, performance gain is eligable,
by using modded openxr toolkit and crop2fov feature, an aggrassive up and down reduction of image, i can use 120hz and mostly ultra settings with 8x msaa.
I just got crop2fov mod up and running yesterday, those are some impressive numbers! What are your system specs and what %'s are you cropping?

Thanks
 
14600 with rtx5080.
Psvr 2, %100 resolution steamvr.
In game settings 8x msaa, pp high, all settings ultra except shadows-reflections on high and oppenents on medium.
Crop up 50, down 70.
Mostly driving on tracks solo, always time trailing myself.
 
Back
Top