community advice Dynamic Foveated Rendering - Everything in one page

[Dentedend10]

On 2/6/2025 at 4:41 PM, mbucchia said:

This is all explained here: 

 

Hi,

I’d like to thank you again for making our VR dreams come closer to reality with your Quadviews implementation. 
It blows my mind and frustrates me that game developers are still not catching on. Releasing new MSFS, or even ACEvo without this kind of support for high-end VR users. 
I also appreciate ED and DCS for having this kind of compatibility. 
I know you have stopped working on this. But is there a way where we can still thank you for your developments? Looking for a way to donate. 
thank you again!

[Qcumber]

Sorry if this has come up before in this thread but it is a long one. XRNS does not work properly with QVFR. When I look over my shoulder I only see the low res part of the image. Is there any workaround for this? 

[speed-of-heat]

was working just fine 10 mins ago ?

[Qcumber]

3 hours ago, speed-of-heat said:

was working just fine 10 mins ago ?

XRNS works but looking backwards is rendered in the peripheral resolution and does not support DFR. All I see when looking over my shoulder (when XRNS is active) is a blurred image. 

[speed-of-heat]

Errr yes it does...

Work that is

I see a perfectly clear image using DFR and xrns when looking behind 

[Qcumber]

27 minutes ago, speed-of-heat said:

Errr yes it does...

Work that is

I see a perfectly clear image using DFR and xrns when looking behind 

Thanks. I see you are using a Varjo headset. Maybe it's a Meta issue. I guess I will stick with the regular neck and back exercises!

[speed-of-heat]

Maybe, but, then I cant see why it would work in x percent of the arc, is it possible, you are looking sideways at the back and maybe the meta headset doesn't pick up your eyes that way? Try increasing the amount of deflection in xrns perhaps?

[Qcumber]

4 minutes ago, speed-of-heat said:

Maybe, but, then I cant see why it would work in x percent of the arc, is it possible, you are looking sideways at the back and maybe the meta headset doesn't pick up your eyes that way? Try increasing the amount of deflection in xrns perhaps?

Cheers. I'll play with the settings. 

[Qcumber]

9 hours ago, speed-of-heat said:

Maybe, but, then I cant see why it would work in x percent of the arc, is it possible, you are looking sideways at the back and maybe the meta headset doesn't pick up your eyes that way? Try increasing the amount of deflection in xrns perhaps?

I reset the arc to default and it works fine now. Thanks for the tip. 

[TimSell75]

During the recent MRtV Event with the Crystal Super it was mentioned that @mbucchia was able to fix/erase the visible Box when using Quad Views. Can somebody confirm this? Because I still see it.

At around Minute 28 in this video

 

 

[mbucchia]

On 3/29/2025 at 1:55 PM, Qcumber said:

Thanks. I see you are using a Varjo headset. Maybe it's a Meta issue. I guess I will stick with the regular neck and back exercises!

Check the order of the API layers with Fred's tool. I recall XRNS needed to be in a specific order.

On 4/1/2025 at 12:41 AM, TimSell75 said:

During the recent MRtV Event with the Crystal Super it was mentioned that @mbucchia was able to fix/erase the visible Box when using Quad Views. Can somebody confirm this? Because I still see it.

At around Minute 28 in this video

 

 

I have zero clue what he is talking about. ED resolved (or greatly improved) that issue, I can't remember when but it wasn't so long ago. I have no hand into DCS rendering or engine.

Make sure you also disable the couple of options listed in the QVFR guide.

Edited April 3 by mbucchia

[Xaliance]

Hello,

I have a Pimax Crystal Light and i have the same problem with the box being visible when using quadview .

The box  is only visible in the left eye, even tough i followed the instructions carefully.

 

Ps : I also tried Pavlov VR to see if it was due to DCS , and this time the box is visible in both eyes

 

[speed-of-heat]

lowering the brightness will decrease the visibility of the effect

 

[kerlcat]

On 4/3/2025 at 5:14 PM, Xaliance said:

Hello,

I have a Pimax Crystal Light and i have the same problem with the box being visible when using quadview .

The box  is only visible in the left eye, even tough i followed the instructions carefully.

 

Ps : I also tried Pavlov VR to see if it was due to DCS , and this time the box is visible in both eyes

 

I have same issue with Super. A bit annoying.

[Qcumber]

I have noticed today that the foveated region is not as clear and sharp as the image is without QVFR at the same resolution. For example, if I use VD at Godlike resolution the image looks clear and sharp without applying any sharpening. With QVFR set to 1.0x centre and 0.5x periphery (0.25x0.25 size) the foveated image is "fuzzy". I have to apply quite a lot of sharpening and it still does not look as good. 

[DLSS Quality, 72Hz, latest version of VD, HEVC 10bit at 150mbps]

[mbucchia]

4 hours ago, Qcumber said:

I have noticed today that the foveated region is not as clear and sharp as the image is without QVFR at the same resolution. For example, if I use VD at Godlike resolution the image looks clear and sharp without applying any sharpening. With QVFR set to 1.0x centre and 0.5x periphery (0.25x0.25 size) the foveated image is "fuzzy". I have to apply quite a lot of sharpening and it still does not look as good. 

[DLSS Quality, 72Hz, latest version of VD, HEVC 10bit at 150mbps]

The math QVFR does is to make the final stereo image at the resolution of what the focus region (inner view) would be if it covered the full FOV.

So the final stereo image is 1:1 the same resolution that you have without QVFR if you set the focus multiplier to 1.0, and the pixels in that inner view are 1:1 the same resolution.

I would try comparing without DLSS, because I dont know what the engine does when it has to do quad  DLSS instead of stereo. It's quite possible that the "1:1 resolution" is thrown away by the engine (which isn't something I have any control on, and would be an engine bug IMO).

You can perhaps post the log file so we can compare the resolution QVFR sets. This won't tell us what resolution DLSS uses though.

Edited May 25 by mbucchia

[Qcumber]

3 hours ago, mbucchia said:

The math QVFR does is to make the final stereo image at the resolution of what the focus region (inner view) would be if it covered the full FOV.

So the final stereo image is 1:1 the same resolution that you have without QVFR if you set the focus multiplier to 1.0, and the pixels in that inner view are 1:1 the same resolution.

I would try comparing without DLSS, because I dont know what the engine does when it has to do quad  DLSS instead of stereo. It's quite possible that the "1:1 resolution" is thrown away by the engine (which isn't something I have any control on, and would be an engine bug IMO).

You can perhaps post the log file so we can compare the resolution QVFR sets. This won't tell us what resolution DLSS uses though.

Thanks for the quick reply. Here are log files for DLSS Quality and no AA at foveated 1.1 (settings as above). If I push the foveated region to 1.8 (I know, this sounds crazy) it looks great in DLSS and there is no ghosting. For some reason it does not want to log the QVFR 1.8 run with DLSS. Not sure why!?

I should add that I am not complaining about QVFR. It has made a massive difference for me in DCS even since I upgraded to the 5080. 

 

 

Quad-Views-Foveated-no AA.log Quad-Views-Foveated-DLSS.log Quad-Views-Foveated-no AA-res 1.8.log

[mbucchia]

3 hours ago, Qcumber said:

For some reason it does not want to log the QVFR 1.8 run with DLSS. Not sure why!?

The log just repeats the content of your config. So there's likely a user error here.

 

For the rest you can do the math.

With your focus size settings at 0.25 on both axes, it means the resolution of the focus region will be 1/4th of your stereo resolution (Godlike).

3072 / 4 = 768

Because you are using 1.1 multiplier, we need to also factor that

768 × 1.1 = 844

Which is the resolution we see being advertised to the game (in the log file).

 

Now to look at the actual render resolution with DLSS, at quality this means input is 0.67 factor 

Stereo: 3072 × 0.67 = 2058

QVFR focus: 844 × 0.67 = 565

That's assuming the game is actually following the same math. One would need to enable the DLSS logging in order to check that.

[Qcumber]

19 hours ago, mbucchia said:

The log just repeats the content of your config. So there's likely a user error here.

 

For the rest you can do the math.

With your focus size settings at 0.25 on both axes, it means the resolution of the focus region will be 1/4th of your stereo resolution (Godlike).

3072 / 4 = 768

Because you are using 1.1 multiplier, we need to also factor that

768 × 1.1 = 844

Which is the resolution we see being advertised to the game (in the log file).

 

Now to look at the actual render resolution with DLSS, at quality this means input is 0.67 factor 

Stereo: 3072 × 0.67 = 2058

QVFR focus: 844 × 0.67 = 565

That's assuming the game is actually following the same math. One would need to enable the DLSS logging in order to check that.

Godlike resolution is 3072x3216 so this would equate to 9.9 pixels full image (monocular). 

Foveated region: 0.25x0.25 = 0.0625 or 6.25% of the overall image size or 6.2 Mp pixels. With 1.1 multiplier this would be 0.68 Mp pixels. 

Peripheral region: 100-6.25 = 93.75% of the total image. ((9.9 x 0.9375) x 0.5 = 4.6 Mp

Total: 0.68 + 4.6 = 5.3 Mp

QVFR vs no QVFR: 9.9 Mp vs 5.3 Mp, about half. 

DLSS: (x0.67) 6.6 vs 3.6. 

In theory QVFR should give a big increase in performance and it does. However, the image quality in the centre is worse than if no QVFR is applied. Even with no AA.

I need to apply a 1.8 foveated multiplier to get similar quality to Godlike and no QVFR. Using the maths above this would mean that the foveated region would be 1.1 Mp. And the total of 5.7 Mp (3.8 with DLSS). Still a big difference to 9.9 (6.6). 

I am curious as to why this extra multiplier is needed to get the same image quality.  

Again I just want to emphasize I am not criticizing QVFR.

 

 

Edited May 26 by Qcumber

[Silver_]

Personally, with my Quest 3 and QVFR I get a much higher quality than VD's Godlike mode, of course I'm using DLSS and wired link and I only use the Oculus Tray tool to change the sound from my glasses to the computer and vice versa. That said, with my 3090, I can't use mode J and K because I notice a drop in performance. On the other hand, I use mode E which gives me higher image contrast and very good performance (72 frames in most situations), with some ghosting (very little) but I've gotten used to it. I have the fovetted region adjusted so that my eye doesn't see the box with the naked eye, I have to look very far to the edge of the lens to see it.

[mbucchia]

6 hours ago, Qcumber said:

Godlike resolution is 3072x3216 so this would equate to 9.9 pixels full image (monocular). 

Foveated region: 0.25x0.25 = 0.0625 or 6.25% of the overall image size or 6.2 Mp pixels. With 1.1 multiplier this would be 0.68 Mp pixels. 

Peripheral region: 100-6.25 = 93.75% of the total image. ((9.9 x 0.9375) x 0.5 = 4.6 Mp

Total: 0.68 + 4.6 = 5.3 Mp

QVFR vs no QVFR: 9.9 Mp vs 5.3 Mp, about half. 

DLSS: (x0.67) 6.6 vs 3.6. 

In theory QVFR should give a big increase in performance and it does. However, the image quality in the centre is worse than if no QVFR is applied. Even with no AA.

I need to apply a 1.8 foveated multiplier to get similar quality to Godlike and no QVFR. Using the maths above this would mean that the foveated region would be 1.1 Mp. And the total of 5.7 Mp (3.8 with DLSS). Still a big difference to 9.9 (6.6). 

I am curious as to why this extra multiplier is needed to get the same image quality.  

Again I just want to emphasize I am not criticizing QVFR.

 

 

I dont follow your math. See the log file for the megapixels math (it's done there) but also this megapixels is irrelevant to quality, since what matters is pixel density. 

QVFR tells the game to preserve 1:1 pixel density in the focus region (well, with your multiplier, 1.1x pixel density). We confirmed that in your log.

If the game doesn't render at this density, then it's a game bug.

Someone probably needs to do a deeper investigation in order to confirm.

 

Edit: can you also look at your VD overlay? It should say something like resolution = 110%

Edited May 27 by mbucchia

[EX_AI]

@mbucchia, thank you very much for your Quad-Views-Foveated, but I think I'm in big trouble now.

When I used Quad-Views-Foveated on my 4060ti 16G, it gave me about a 20% performance boost at ultra settings with 120% center resolution.
This plugin helped me a lot until I upgraded to a 5070ti: I was very excited to use Quad-Views-Foveated again, but the results were terrible: enabling it actually caused my frame rate to drop—by an average of 18% with the default configuration. Moreover, Quad-Views-Foveated itself caused blurring in the center of the view (as mentioned in a post, which I’m sure you’re aware of). In the end, I got very disappointing results: with Quad-Views-Foveated enabled, both image quality and frame rate significantly decreased—a complete negative gain.

I’ve tried many methods and spent a lot of time attempting to fix this, including reinstalling the game, reinstalling the plugin, and even reinstalling the system, but nothing worked—the problem persists...

I sincerely request your help... please help me.

Below are my settings screenshots and log files. The test map was: low-altitude free flight over the Mariana Islands in the F16C.

Quad-Views-Foveated.log

From the 4060 Ti to the 5070 Ti, the plugin's effect changed from positive to negative—is all this related to the GPU upgrade? Or did I misunderstand something?

Edited June 3 by EX_AI

[EX_AI]

I tried again—this time setting the center region to 0.3x0.3. Under the same test scenario, the performance was 54-56 FPS, still not reaching the frame rate when quad views are disabled（56-60FPS), and the visual quality was severely degraded (Quest 3 doesn’t have eye tracking).

And the core issue probably isn’t the center region, because I previously used this exact same configuration on a 4060 Ti, where it gave me an 8 FPS boost.Quad-Views-Foveated (1).log

 

Additionally, I observed that with the plugin enabled during runtime, my CPU usage stays around 80% while GPU usage remains at approximately 96%. Although the in-game visuals on the PC remain smooth at this point, switching to other applications (such as Task Manager or a browser) causes severe stuttering—a single mouse click may take several seconds to register any response.

For comparison, when the plugin is disabled, CPU usage drops to around 62% (GPU usage remains at ~96%), and no significant stuttering occurs when switching between applications. I'm not sure if describing these observations will help pinpoint the root cause of the issue, but I’m providing the details in case they’re useful.