VR experiences IL-2 Korea

[AEthelraedUnraed]

3 hours ago, Aapje said:

There are huge differences in expertise between programmers. I suggested before that they hire mbucchia (the maker of OpenXR Toolkit) as a consultant, because he is clearly world class when it comes to VR. No one at 1CGS will even have the opportunity to spend that much time on VR, nor did they ever work at a company making VR hardware, so they don't/didn't have the opportunity to get that much expertise.

I'd say that there is a pretty good chance that he would be able to point out ways to improve VR in the game beyond what 1CGS can come up with, whether that is with a better foveated rendering implementation or other improvements.

Well yes, if you hire an expert you're going to achieve better results 🙂

Still, I think it's a stretch to say that their experimentally achieved figure of only a 10% speedup is false because they didn't hire one. Fact is, no-one except the Devs knows how Korea's rendering pipeline works, where its bottlenecks are, and whether or not foveated rendering would work well. Since it's their money on the line, I'm going to assume they did a proper investigation and didn't put a complete n00b programmer on it but rather an established member of their team.

36 minutes ago, Antonovomatic said:

You're making the same mistake the author of that answer to the question seems to have, conflating variable rate shading and quad-views

No, I'm not. No matter which technique you use for foveated rendering, you need to draw the geometry for the whole screen. If there are a million vertices, you still need to run your calculations on a million vertices whether if they're in the low-res or high-res area. What mbuchchia's saying is that by re-using some parts of the pipeline, you don't need to fully render all of that again resulting in relatively cheap rendering for the high-res part as you've already done some of that work. Using quad-view rendering, you might save a lot in shading, rasterisation etc. since you're rendering fewer pixels overall, but if the rendering costs in Korea are heavily geared towards the geometry stage, the net gain might be disappointing.

[Dash,Polder]

https://diener.tech/products/vectorxr

You can't use the quad views but the other three things work

When you select a new device, refresh first before you select and than pin a button. 

Edited Saturday at 06:42 PM by Dash,Polder

[YoYo]

Since OpenXR Toolkit is having issues (CTD due to the new OpenXR runtime), has anyone tried Reshade VR? For example, it also has an option to sharpen the image in VR (CAS), which always improves visibility.

https://reshade.me/

[Antonovomatic]

1 hour ago, AEthelraedUnraed said:

Well yes, if you hire an expert you're going to achieve better results 🙂

Still, I think it's a stretch to say that their experimentally achieved figure of only a 10% speedup is false because they didn't hire one. Fact is, no-one except the Devs knows how Korea's rendering pipeline works, where its bottlenecks are, and whether or not foveated rendering would work well. Since it's their money on the line, I'm going to assume they did a proper investigation and didn't put a complete n00b programmer on it but rather an established member of their team.

No, I'm not. No matter which technique you use for foveated rendering, you need to draw the geometry for the whole screen. If there are a million vertices, you still need to run your calculations on a million vertices whether if they're in the low-res or high-res area. What mbuchchia's saying is that by re-using some parts of the pipeline, you don't need to fully render all of that again resulting in relatively cheap rendering for the high-res part as you've already done some of that work. Using quad-view rendering, you might save a lot in shading, rasterisation etc. since you're rendering fewer pixels overall, but if the rendering costs in Korea are heavily geared towards the geometry stage, the net gain might be disappointing.

If geometry was such a significant portion of the workload there wouldn't be such disparity in performance between flatscreen and VR that many are reporting here.  It would also incur massive swings in workload when changing FOV/zoom (and therefore viewing frustrum).

Nothing in the dev answer to the question of foveated rendering mentions or implies such tests involved quadview rendering and we should not assume that greater performance gains than the stated <10% would not be possible, unless the goal is to arbitrarily discourage any investigation of the technology.

Edit: The simple fact is that users of top-end HMDs can, want to, and are pushing way more pixels than just a few years ago, and outside of upscaling tech (which I'm optimistic about and stoked to see supported already in Korea) or raw GPU horsepower, foveated rendering techniques are the only way to maintain some sort of performance parity at those ultra-high resolutions.

Edited Saturday at 07:54 PM by Antonovomatic

[AEthelraedUnraed]

25 minutes ago, Antonovomatic said:

If geometry was such a significant portion of the workload there wouldn't be such disparity in performance between flatscreen and VR that many are reporting here.  It would also incur massive swings in workload when changing FOV/zoom (and therefore viewing frustrum).

You do make some good points there (although I personally haven't witnessed the huge disparity in performance; my VR works just fine).

Still, I feel we shouldn't discount the Devs' investigation. We don't know how their rendering pipeline looks like, where and how a foveated rendering technique (whichever) would fit, and what effect that would have. Any statement that it needs to be more than 10% is based on the performance of other games with other rendering pipelines.

30 minutes ago, Antonovomatic said:

Edit: The simple fact is that users of top-end HMDs can, want to, and are pushing way more pixels than just a few years ago, and outside of upscaling tech (which I'm optimistic about and stoked to see supported already in Korea) or raw GPU horsepower, foveated rendering techniques are the only way to maintain some sort of performance parity at those ultra-high resolutions.

True. Although that begs the question what percentage of their users would actually benefit from such a change. I know I wouldn't with my Quest2, and even though eye tracking is getting more common, I suspect that's still the case for the majority of VR users. And based on polls on other flight sims, VR users are probably a tiny minority. Even if they'd get (much) more than the stated 10% performance increase, is it financially worth it just to silence that (very vocal) minority?

[YoYo]

7 minutes ago, AEthelraedUnraed said:

actually benefit from such a change

I have a Quest Pro with active eye tracking. I tested DFR in DCS and MSFS, but it was completely useless for me. It had no effect (a 5-10% increase in FPS, if any) and caused artifacts around the sharp edge of the field, which could be seen out of the corner of my eye. The result was that I didn't see any benefit, and I still had flickering (a lower-resolution image around it). I disabled it and don't use it anymore, although I know the people at Pimax like it and sometimes it's the only solution for them to improve performance.

[Antonovomatic]

1 hour ago, YoYo said:

I have a Quest Pro with active eye tracking. I tested DFR in DCS and MSFS, but it was completely useless for me. It had no effect (a 5-10% increase in FPS, if any) and caused artifacts around the sharp edge of the field, which could be seen out of the corner of my eye. The result was that I didn't see any benefit, and I still had flickering (a lower-resolution image around it). I disabled it and don't use it anymore, although I know the people at Pimax like it and sometimes it's the only solution for them to improve performance.

You should have been getting comparable results to enforcing FOV limits in the Oculus debug panel, was this OXRToolkit or Quadviews that you tried?  I run QV w/the foveated section at 45% x 45% FOV so I rarely see the borders at 110% scaling for some clarity headroom, with peripheral res at 25% and a blend/transition of 25.  According to Vector XR that results in ~31% rendered pixel count and the in-game performance reflects that massively reduced GPU workload.

Edited Saturday at 09:53 PM by Antonovomatic

[Jade_Monkey]

1 hour ago, AEthelraedUnraed said:

Even if they'd get (much) more than the stated 10% performance increase, is it financially worth it just to silence that (very vocal) minority?

What's the point of promoting this as an upgraded engine in all the industry events with a VR station for people to try, and then releasing an unpolished wreck that only works in old headsets your grandma used and does not work well with anything new? including higher resolution and eye tracking functionality that has benefited all other major sims?

 

Edit: btw, I'm ok with them saying that they have a few big tickets items in their list they need to work on (e.g. career, dlc) and this has not been the top priority. But so far it sounds like they have dismissed these solutions after that 10% test.

Edited Saturday at 10:20 PM by Jade_Monkey

[David]

On my  Ryzen 9 9950X3D with 64GB DDR5 RAM and a RTX 5090, The game looked like a slide show of boiled buttocks at 15fps on my Pimax Super Ultrawide.

I had basically copied my DCS Pimax settings as a base setting, but I had to drop the render to auto and use the NIS upscaling to get a steady locked 40 to 45 fps on high game settings. The upscaling makes the cockpit look like a fifteen year old DCS FC3 module, but at least it's smooth(ish) and playable, but they don't seem to have learned much from previous iterations of il2, because VR as always feels like an afterthought.

It's early days I suppose and still in EA, so I'll give the game a chance to improve. 

[Aapje]

7 hours ago, AEthelraedUnraed said:

Well yes, if you hire an expert you're going to achieve better results 🙂

Still, I think it's a stretch to say that their experimentally achieved figure of only a 10% speedup is false because they didn't hire one. Fact is, no-one except the Devs knows how Korea's rendering pipeline works, where its bottlenecks are, and whether or not foveated rendering would work well.

The reason why IL-2 can do both flat-screen and VR without requiring two separate games is because DirectX is quite open, and OpenVR and OpenXR integrate with it, to send the rendered images to the headset driver (which has to modify them to work with the headset). This openness means that it is actually possible for someone who doesn't have access to the code of the game, to see what happens in DirectX and in OpenXR, and even to modify things. So it is not in fact true that only 1CGS can know how the rendering pipeline works. An experienced DirectX coder can investigate that as well.

This is actually why OpenXR Toolkit could exist in the first place, and also why Reshade works and why people have been able to turn flat screen games into VR games, without having the source code of the game.

And mbucchia has previously done a review of how MSFS does its rendering for VR, and found issues with it, without access to the code, and published that on the MSFS forums.

So if he was willing to do that for free, I would expect him to gladly help 1CGS for a modest fee and be able to do an investigation quite quickly. 

[MiGCap]

I have a standard Pimax (the one which came on the market first before the Super and the Light) and it works well with my plain vanilla 3090.

I have relatively high quality settings and have little problems with my 35 FPS, even when they drop in the 20s over cities, as the game runs much smoother on these low FPS than in DCS (where I have 10 FPS more averagely in max settings in near all areas). It is only noticable when You look out at the side of the cockpit in low altitude flight. So why do You need 70 FPS and more? Even in IL-2 GB I have "only" around 50 FPS on max settings and I am happy with the quality. I didn't even change my VR settings when I do my screenshots from tracks in 2D.

I am very happy with the VR experience so early in the game and I am sure it will be developed even further in the future.

My current settings, flown in MP and SP (but no B-29 intercept until now, so I can't give a statement about the performance in this case).

 

[MoleUK]

9 hours ago, YoYo said:

I have a Quest Pro with active eye tracking. I tested DFR in DCS and MSFS, but it was completely useless for me. It had no effect (a 5-10% increase in FPS, if any) and caused artifacts around the sharp edge of the field, which could be seen out of the corner of my eye. The result was that I didn't see any benefit, and I still had flickering (a lower-resolution image around it). I disabled it and don't use it anymore, although I know the people at Pimax like it and sometimes it's the only solution for them to improve performance.

Also using a QPro here.

Are you talking about DFR or Quad views?

With Quad views in DCS you can see a 30-50+% performance increase depending on how aggressive you are with the tuning.

With DFR injected via OpenXRToolkit as an example, the gains are far far less significant. Though it doesn't tax the CPU like quad views does.

The artifacting is a problem with both methods, some find it too distracting. But I literally couldn't run DCS at anywhere near the resolution I do without quad views.

I can't speak to MSFS performance.

Edited yesterday at 05:50 AM by MoleUK

[YoYo]

1 hour ago, MoleUK said:

Also using a QPro here.

Are you talking about DFR or Quad views?

With Quad views in DCS you can see a 30-50+% performance increase depending on how aggressive you are with the tuning.

With DFR injected via OpenXRToolkit as an example, the gains are far far less significant. Though it doesn't tax the CPU like quad views does.

The artifacting is a problem with both methods, some find it too distracting. But I literally couldn't run DCS at anywhere near the resolution I do without quad views.

I can't speak to MSFS performance.

In this case, as I wrote. I wrote about improving performance thanks to Dynamic Foveated Rendering. And as I wrote, for me, there was essentially no increase (72Hz and a constant 72FPS without it and without MR). It's possible that both sims simply run so well on my 5090 that I didn't notice any increase, only problematic artifacts. On the other hand, this demonstrates the vast gulf between cities in DCS and IL-2 Korea, in favor of the former's engine. I hope 1CGS will deal with the optimization problem and improve it.

Btw. QV can actually be quite a small increase (as devs noted up to 10%). DFR may give perhaps better growth:

But I don't use QV myself, so I don't base my opinion on my own experience.

Edited yesterday at 07:37 AM by YoYo

[YoYo]

4 hours ago, Aapje said:

So if he was willing to do that for free, I would expect him to gladly help 1CGS for a modest fee and be able to do an investigation quite quickly. 

It seems he needs to be pulled away from Valve, though, as he's currently working there on a VR program called Steam VR/Frame. 

Of course, I keep my fingers crossed and it would be good if 1CGS decided to take this step.

[YoYo]

3 hours ago, MiGCap said:

35 FPS, even when they drop in the 20s over cities, as the game runs much smoother on these low FPS than in DCS (where I have 10 FPS more averagely in max settings in near all areas). It is only noticable when You look out at the side of the cockpit in low altitude flight. So why do You need 70 FPS and more?

I admire that 30 FPS is good for you, even with such terrible optical settings . But it's like hearing and music; some are audiophiles who pay attention to the entire sound range and quality in every frequency range, while others are satisfied with the quality of a mobile radio. There's no accounting for taste, and if that suits you, then of course, great for you 👍. Personally, I see a huge and tangible difference, even between 68 and 72 frames and every shake in the frames, every instability is very visible. Regards.

[Antonovomatic]

37 minutes ago, YoYo said:

In this case, as I wrote. I wrote about improving performance thanks to Dynamic Foveated Rendering. And as I wrote, for me, there was essentially no increase (72Hz and a constant 72FPS without it and without MR). It's possible that both sims simply run so well on my 5090 that I didn't notice any increase, only problematic artifacts. On the other hand, this demonstrates the vast gulf between cities in DCS and IL-2 Korea, in favor of the former's engine. I hope 1CGS will deal with the optimization problem and improve it.

"Dynamic Foveated Rendering" is a general term like "rendered shadows" and there are multiple independent ways to implement it with wildly different performance results.  Both those sims you mentioned are compatible with both methods so it's pertinent to know which you tried, since the improvements from OXRToolkit/"Variable rate shading" are generally miniscule compared to QV.  It really doesn't square that enforcing 0.75 FOV would net you positive results but anything but the most pixel-liberal settings in QV wouldn't (because there's a massive overlap in "the how and why" both reduce workload).

The other major benefit of QV is that when eye-tracked it is significantly less perceptible than the IMO, very obvious rendering zones of OXRTK's implementation (which I must remind everyone again, is abandoned and explained to be far inferior by mbucchia himself) is what makes me think that's what you are referring to

Edited yesterday at 07:50 AM by Antonovomatic

[YoYo]

2 minutes ago, Antonovomatic said:

"Dynamic Foveated Rendering" is a general term like "rendered shadows" and there are multiple independent ways to implement it with wildly different performance results.  Both those sims you mentioned are compatible with both methods so it's pertinent to know which you tried, since the improvements from OXRToolkit are generally miniscule compared to QV.  It really doesn't square that enforcing 0.75 FOV would net you positive results but anything but the most pixel-liberal settings in QV wouldn't (because there's a massive overlap in "the how and why" both reduce workload).

Yes, I fly with a 0.75 FOV, which gives me excellent results (thankfully, it's possible to use it). Below that value, I can already see the frame, but with my goggles, 0.75 is perfect, which saves me a bit on power. 👍

[MoleUK]

43 minutes ago, YoYo said:

In this case, as I wrote. I wrote about improving performance thanks to Dynamic Foveated Rendering. And as I wrote, for me, there was essentially no increase (72Hz and a constant 72FPS without it and without MR). It's possible that both sims simply run so well on my 5090 that I didn't notice any increase, only problematic artifacts. On the other hand, this demonstrates the vast gulf between cities in DCS and IL-2 Korea, in favor of the former's engine. I hope 1CGS will deal with the optimization problem and improve it.

Btw. QV can actually be quite a small increase (as devs noted up to 10%). DFR may give perhaps better growth:

But I don't use QV myself, so I don't base my opinion on my own experience.

Ok, just to be clear that quad views is a form of DFR. It uses quite a different technique, but people often confuse the two. Quad views works quite differently from DFR, but it is a form of DFR.

Quad views tends to lead to far greater performance increases on the GPU side. Your screenshot there is also incorrect, quadviews can/does track eye movement. I use it on the QPro with eyetracking, it tracks eye movement as your gaze moves around and keeps that viewpoint at a higher resolution.

It doesn't require eye tracking, as you can use it in a fixed format. But that is also true with DFR, you can use that without eyetracking too. Both are far less effective without eye tracking since without eye tracking it's far more visually noticeable.

In no circumstance does DFR lead to a greater impact on GPU performance than Quad Views. However DFR does not require more CPU load.

Quad views requires a specific implementation by the developer, very few games have implemented it. DCS, Pavlov MSFS and iRacing are the only titles that have implemented quad views afaik. It also won't work on Vulkan without a translation layer of sorts, so quad views may stop working with DCS at some point unless they take steps to address it.

DFR you can simply use with any OpenXR title via OpenXRtoolkit. You can probably use it with Korea right now, I haven't tried as of yet.

It sounds like you tried DFR with DCS etc rather than Quad views. And yes DFR doesn't lead to a massive performance improvement like quadview can. It's still nice though.

Neither DFR or Quad Views are too useful if you just enable it without changing settings, at least if you're already hitting your refresh rate. The idea is to increase your available GPU overhead so you can either raise the resolution or raise settings and still be able to maintain 72hz.

Edited yesterday at 07:59 AM by MoleUK

[MoleUK]

14 minutes ago, Antonovomatic said:

"Dynamic Foveated Rendering" is a general term like "rendered shadows" and there are multiple independent ways to implement it with wildly different performance results.  Both those sims you mentioned are compatible with both methods so it's pertinent to know which you tried, since the improvements from OXRToolkit/"Variable rate shading" are generally miniscule compared to QV.  It really doesn't square that enforcing 0.75 FOV would net you positive results but anything but the most pixel-liberal settings in QV wouldn't (because there's a massive overlap in "the how and why" both reduce workload).

The other major benefit of QV is that when eye-tracked it is significantly less perceptible than the IMO, very obvious rendering zones of OXRTK's implementation (which I must remind everyone again, is abandoned and explained to be far inferior by mbucchia himself)

Yeah unfortunately people use the terms DFR and Quad Views interchangeably which does lead to some confusion.

I do wonder if the Q&A made the same mistake and was referring to DFR rather than Quad views. But that wouldn't make too much sense since DFR doesn't need to be directly implemented by the developers, whereas Quad views does.

Edited yesterday at 07:54 AM by MoleUK

[Antonovomatic]

47 minutes ago, MoleUK said:

Yeah unfortunately people use the terms DFR and Quad Views interchangeably which does lead to some confusion.

I do wonder if the Q&A made the same mistake and was referring to DFR rather than Quad views. But that wouldn't make too much sense since DFR doesn't need to be directly implemented by the developers, whereas Quad views does.

"DFR" accurately describes what QV is/does, just like how a P-51 and B-29 are both "warplanes" but only one is a fighter.  It's Variable Rate Shading that is the term which describes and distinguishes the technique OXRTK uses.
Dynamic - it moves
Foveated - Really a marketing term by the first company to demo the tech, but is basically understood to mean there is a consideration of what is intended/expected to be in central vs peripheral vision
Rendering - It involves the rendering of pixels

[Nikita]

12 hours ago, AEthelraedUnraed said:

Although that begs the question what percentage of their users would actually benefit from such a change.

Maybe the kind of percentage that has the money to buy more DLC in the future lol

[YoYo]

1 hour ago, MoleUK said:

DFR you can simply use with any OpenXR title via OpenXRtoolkit. You can probably use it with Korea right now, I haven't tried as of yet.

Unfortunately not, because IL-2 Korea uses a newer OpenXR runtime than OpenXR Toolkit, which causes CTD in IL-2 Korea. Mbucchia hasn't updated this for about 2 years.

Btw. I used the DFR included with the Toolkit and also tested the static FR.

[AEthelraedUnraed]

7 hours ago, Aapje said:

So it is not in fact true that only 1CGS can know how the rendering pipeline works. An experienced DirectX coder can investigate that as well.

I said "knows", not "can know" 🙂. Even without DirectX, there's always decompilation so unless the code is heavily obfuscated or something, it's never impossible to figure out what happens. But AFAIK no-one has done so, so my point stands.

1 hour ago, Nikita said:

Maybe the kind of percentage that has the money to buy more DLC in the future lol

I hope you're not suggesting that DLCs should be something for the rich?

According to the latest Navigraph Survey, around 10% of players primarily fly VR (this is geared towards flight sims in general, but it's probably the best stats we've got). Based on the Steam Hardware & Software Survey, a maximum of 10.61% of all headsets feature eye-tracking (every ambiguity counted as eye-tracking - in reality it's likely *much* less). That means that perhaps 1% of all IL2 users would benefit from this. Even if this 1% are all going to buy all upcoming DLCs (which they won't), even if they'd all stop buying DLCs until foveated rendering is implemented (which they wouldn't), I doubt this "percentage that has the money to buy more DLC in the future" has that much of an influence on 1CGS's total revenue to make it worth the investment at this point in time.

Edited yesterday at 10:21 AM by AEthelraedUnraed

[MoleUK]

1 hour ago, Antonovomatic said:

"DFR" accurately describes what QV is/does, just like how a P-51 and B-29 are both "warplanes" but only one is a fighter.  It's Variable Rate Shading that is the term which describes and distinguishes the technique OXRTK uses.
Dynamic - it moves
Foveated - Really a marketing term by the first company to demo the tech, but is basically understood to mean there is a consideration of what is intended/expected to be in central vs peripheral vision
Rendering - It involves the rendering of pixels

Yeah it just unfortunately causes confusion. And now valve/VD has dynamic foveated streaming just to further confuse people lol.

1 hour ago, YoYo said:

Unfortunately not, because IL-2 Korea uses a newer OpenXR runtime than OpenXR Toolkit, which causes CTD in IL-2 Korea. Mbucchia hasn't updated this for about 2 years.

Btw. I used the DFR included with the Toolkit and also tested the static FR.

That's a shame about the runtime. I know someone else is working on an alternative toolkit but I think it's more for quad views and necksafer atm.

Quad-views is a seperate app/install that isn't included in the OpenXRtoolkit. The developer who made OpenXRtookit did also make quad views.

If you ever do want to give quad views a try, you can install it here. But as I said unfortunately only a couple of VR titles support it currently.

https://github.com/mbucchia/Quad-Views-Foveated

Having it installed can also cause some issues with UEVR games iirc, you have to specifically whitelist them or uninstall it at that point.

Edited yesterday at 10:41 AM by MoleUK

[pocketchange]

On 6/27/2026 at 8:57 AM, brickcommander said:

I am also however very annoyed with the pilot head limitation. Turning it off doesnt work and every small yead movement- particularly up/down- moves the whole cockpit around me. And the menu screens are too close and it is uncomfortable. I would imagine both issues are easy fixes that will he resolved soon.

Just a heads up - this was fixed for me when I recalibrated my VR space to seat only in SteamVR. It is a little disappointing that apparently I'll have to do this every time I switch, but this and a 115% world scale makes an insane difference for cockpit immersion.

I also played around with the graphics settings a lot today, and on my bigscreen beyond 2 + 9070XT setup the only options that actually affect GPU performance seem to be the general presets (anything other than low at native resolution ends up with horrible frametimes, ~19ms) and VR Resolution Scale. I'm at 75% Resoltution Scale on the Balanced Preset now and actually get the smooth 75Hz I need.

As for Upscaling - it flat out doesn't work, but I've also never seen it used in a VR Game? Wouldn't it look quite bad, as any temporal effect would with the rapid head movements VR Displays do? If it does work it would be a great fix for full release since resolution really is the main performance drain on GPU.