DCS VR Performance Optimization

[evanswillo]

Yes this is a beta driver https://developer.download.nvidia.com/compute/cuda/12.8.0/local_installers/cuda_12.8.0_571.96_windows.exe

 

[jurinko]

I have finally reached VR nirvana with my Quest 3, cable link.

1. Use Meta quest link as openxr runtime and in openxr toolkit set Turbo mode on. Without Turbo it stutters. I ran dcs with SteamVR runtime before, it was acceptable but with PD of 1.2 max and ASW did not worked. With Meta/Turbo i can use 1.4 (Meta resolution 1.0/72Hz).

2. I had been using H265 codec at 150 bitrate from some reason (above 200 it gave me fps 1). With H264 and bitrate 500, the feeling is like ramping PD from 1.0 to 1.3.

3. New DLSS4/profile J further improved the visual quality. I put PD back one notch to 1.3 as I pay for it, but it looks good as never before.

4. Switch shadows off in dcs settings, it is broken and consumes plenty of power for nothing.

[yoadknux]

Profile J is an eye candy for sure, allows you to take down PD by 0.1-0.2 and still get good clarity (you can turn up other settings thanks to the lower PD too) but it has bad ghosting. You won't feel it if all you do is fly solo and admire the scenery, but once you start interacting with the enviorement (other planes, missiles, scrolling through MFDs etc) it becomes very noticable. 

[jurinko]

Im lucky I see none. Only if ASW kickes in, is the nearby passing airplane doubled. And new nvidia driver is due this week, promised to supply dlss 4 goodies without performance hit. 

[diamond26]

 

5 hours ago, jurinko said:

Switch shadows off in dcs settings, it is broken and consumes plenty of power for nothing.

That's an interesting statement. What exactly you mean by broken? I use Shadows low or medium and the visual effect is clearly there. 

Edited January 28 by diamond26

[gonvise]

36 minutes ago, diamond26 said:

 

That's an interesting statement. What exactly you mean by broken? I use Shadows low or medium and the visual effect is clearly there. 

I was going to say the same thing, shadows, both in the cockpit and in the world, are clearly visible and one of the things that give the most immersion and graphic quality to this simulator.

[jurinko]

For me it is not worth the voluptuous fps cost. With no shadows, you still get shadows for trees and buildings.

[Ready]

2 hours ago, diamond26 said:

 

That's an interesting statement. What exactly you mean by broken? I use Shadows low or medium and the visual effect is clearly there. 

I have shadows at least set to medium. Really like the effects in the cockpit of Shadow High, but I can also live with Medium. I am seeing a strange effect though on treelines and fences/walls where a part of the shadows only renders when I am pretty close, like 1NM. Will see if I can catch it in a recording

[av8orDave]

I think shadows comes down largely to the details of the system. A few years back, I had a 3080, and I had to be very conscious of the shadows setting, as anything over medium would really kill performance and was a major limiter. I have a 4090 now, and graphics settings really aren't the obstacle... if I get poor framerates, it is almost entirely because large numbers of AI units are killing the CPU (Ryzen 9 7950X). 

Said another way, if I'm flying solo missions with a 4090, I can go hog-wild on graphics settings with little consequence. If I'm flying a complex mission, all bets are off.

Edited January 28 by av8orDave

[jurinko]

5 hours ago, diamond26 said:

 

That's an interesting statement. What exactly you mean by broken? I use Shadows low or medium and the visual effect is clearly there. 

My stupid, I meant ground shadows. Normal shadows are on high.

[Ready]

DLSS Profile J and Profile K

The difference between both profiles is very little in my case. I will probably fly with K until the next improvement is released.

All so beautiful now! Will adjust/lower some QVFR and DLSS settings to trade some clarity for performance. 

Edited February 1 by Ready

[yoadknux]

22 hours ago, Ready said:

DLSS Profile J and Profile K

The difference between both profiles is very little in my case. I will probably fly with K until the next improvement is released.

All so beautiful now! Will adjust/lower some QVFR and DLSS settings to trade some clarity for performance. 

Did you see any ghosting in profile K?

With profile J, clarity is much better compared to profile F, but ghosting is worse. I'm wondering if I should give profile K a chance.

[Ready]

3 hours ago, yoadknux said:

Did you see any ghosting in profile K?

With profile J, clarity is much better compared to profile F, but ghosting is worse. I'm wondering if I should give profile K a chance.

Just try/compare it for yourself I would say. The opinions I read so far are mixed. I might also go back and forth for a while and settle.

Edited February 1 by Ready

[Ready]

Further optimizations

I am playing around with a couple of things at the moment:

1. DCS Object Shadows Flat <> Default.
2. Nvidia DLSS 4.0 Profile J <> K.
3. Combinations of DCS PD (1.0, 0.9 or 08) <> DLSS Performance, Balanced, or Quality <> QVFR Focus_Multiplier (1.0, 1.1 or 1.2)
4. Combinations of Sharpness in DLSS (currently .3) <> Sharpen_Focus_View in QVFR (currently .5)

OpenKneeboard 

1. Single <> Double Kneeboard. Check if there is a performance difference.
2. Only opening the Kneeboard when I need it during flight and keeping it closed otherwise. Having it open costs about 3-4 FPS and a 5-15% hit on the 1% and 0.2% 
3. Reduce amount of images in the Kneeboard to reduce VRAM. (thanks Fred for the tip). 

Quad View Foveated Rendering

1. Changing the dimensions of the focus section. Make them just large enough not to notice the focus section moving around. Currently both at .35 maybe reducing the vertical and increasing the horizontal.

OBS

I usually fly with OBS enabled. So I will start running tests again with OBS enabled.

1. Disabling Desktop Preview. I always used to have the Preview enabled, will start disabling it after I checked that it is working.
2. Adjusting bitrate. Currently possibly set too high.

Edited February 4 by Ready

[actually_fred]

11 hours ago, Ready said:

• Only opening the Kneeboard when I need it during flight and keeping it closed otherwise. Having it open costs about 3-4 FPS and a 5-15% hit on the 1% and 0.2% 

OpenKneeboard prepares the kneeboard images in VRAM regardless of whether or not it's visible - all that showing/hiding it affects is whether or not the result (a single image which is already in VRAM) is passed on to the runtime; this is *extremely* cheap compared to everything the game (and the main openkneeboard process) is doing.

 

If your numbers are reliable, it means either:

- your system is horribly overloaded

- your OpenXR runtime is terribly optimized for composition quad layers, and you should report this as a bug to your OpenXR runtime vendor

> OBS

If you're using Jabbah's OpenXR API layer, that must do additional work when overlays are present (i.e. when OpenKneeboard is visible); I've not looked at how well it works or how well it is optimized. This would be an alternative to 'the runtime is terribly optimized'.

> Reduce amount of images in the Kneeboard to reduce VRAM. (thanks Fred for the tip)

To be clear, this was not a performance tip, just a statement that images stay in VRAM in OpenKneeboard. 'use less VRAM' does not improve performance unless you are reaching 100% full VRAM. When this happens, the game will generally crash or you'll get somewhere between single-digit FPS and 'seconds per frame'

Edited February 4 by actually_fred

[VR Flight Guy in PJ Pants]

I did some benchmarks in VR and compare the difference between no AA with DLSS"4", playing the same replay in VR and here we go:

Surprising the performance in this test between them are very, very close.

Oh, forgot to mention: the video comes shortly.

Edited February 4 by VR Flight Guy in PJ Pants

[diamond26]

I wondered if DCS actually never changed the DLSS option from quality to none as the performance is almost identical which shouldn't be the case. I have the feeling, as I mentioned elsewhere, that Global override keeps the DLSS Quality level at a certain point (in DLSSTweaks it was 0.6666 which is the quality one)

[Ready]

8 hours ago, actually_fred said:

OpenKneeboard prepares the kneeboard images in VRAM regardless of whether or not it's visible - all that showing/hiding it affects is whether or not the result (a single image which is already in VRAM) is passed on to the runtime; this is *extremely* cheap compared to everything the game (and the main openkneeboard process) is doing.

If your numbers are reliable, it means either:

- your system is horribly overloaded

- your OpenXR runtime is terribly optimized for composition quad layers, and you should report this as a bug to your OpenXR runtime vendor

> OBS

If you're using Jabbah's OpenXR API layer, that must do additional work when overlays are present (i.e. when OpenKneeboard is visible); I've not looked at how well it works or how well it is optimized. This would be an alternative to 'the runtime is terribly optimized'.
> Reduce amount of images in the Kneeboard to reduce VRAM. (thanks Fred for the tip)

To be clear, this was not a performance tip, just a statement that images stay in VRAM in OpenKneeboard. 'use less VRAM' does not improve performance unless you are reaching 100% full VRAM. When this happens, the game will generally crash or you'll get somewhere between single-digit FPS and 'seconds per frame'

I don't think my system is horribly overloaded tbh. I believe I still have a lot of headroom in the 9800X3D, even though DCS tells me I am CPU-bound and it somehow has no way to make better use of the CPU's performance. My 4090 is not really doing much during the quick testruns (I start the stock A-10C Marianas take off mission and keep the Hog standing there). I keep my headset on my desk and I sometimes pick it up to look around and check visuals. But the measurements are always done with the headset on the desk and VR centered. During these measurements the VRAM is relatively low I think.

I will check VRAM in future tests when I have removed a bunch of images from the various Tabs in OpenKneeboard. There is about 80% of images I guess I don't not really need or use. Unsure if it would help to clean them up, as you say it does not do anything else than consuming some VRAM and I do like having them at my fingertips, just in case I need something.

I have done a clean Windows10 install a couple of weeks ago and stripped it to its underwear. I am still working on finding more ways to further calm down the OS, but it is pretty silent already. Besides DCS I am running the following applications; Process Lasso, WtabletServicePro+children, VoiceAttack, SRS, OBS, Varjo Process Server+children, Simhaptic+DCSRealistic and OpenKneeboard (in various profiles with one or two kneeboards visible). I am checking these apps one by one to measure what is their impact in this mix. 

With regards to the OpenXR runtime; I am still using Jabbah's since there hasn't been a better known way for me to get OpenKneeboard to appear in my recordings. I will check in with him to see if there is anything to look at. It is unclear if he is still active, our last contact was beginning of 2023 and his last release was about that time, I had his last beta. Unbelievable how time flies. I see our chats on OpenKneeboard date back to 2022 when it was version 0.1.2.180. Where does the time go..

I am also still using Mbuccia's QVFR. Have not tried Varjo's because it won't show OpenKneeboard and I can not manually adjust the peripheral settings for example to get the most out of the system. 

When I don't have OBS running and OpenKneeboard is loaded the difference between hidden and visible shows a difference of a couple of FPS. I will run more tests in various DCS performance settings and configuration states to get a better picture of my system.

(I just noticed that when I close and reopen OpenKneeboard the desktop app does not open in the screen location where it was when I closed it. I also noticed that when I update OpenKneeboard somehow my VR settings like the size, position and orientation of the Virtual Kneeboard gets reset. I will check your help thread and put a few items up on your github if it is not already on there.)

[Ready]

9 hours ago, VR Flight Guy in PJ Pants said:

I did some benchmarks in VR and compare the difference between no AA with DLSS"4", playing the same replay in VR and here we go:

Surprising the performance in this test between them are very, very close.

Oh, forgot to mention: the video comes shortly.

 

 

 

 

 

What is it exactly you are comparing? I will run the same test just to compare if I see the same.

[VR Flight Guy in PJ Pants]

Between with DLSS and no DLSS as shown in the video capture below:

 

 

[Ready]

4 minutes ago, VR Flight Guy in PJ Pants said:

Between with DLSS and no DLSS as shown in the video capture below:

DLSS ON vs DLSS OFF? For me there is a huge difference between the two.

I don't see a video when I click on the image.

[VR Flight Guy in PJ Pants]

1 hour ago, VR Flight Guy in PJ Pants said:

the video capture below

Pardon my poor English, the image was captured from the video. The video is here:

 

[Ready]

10 minutes ago, VR Flight Guy in PJ Pants said:

Pardon my poor English, the image was captured from the video. The video is here:

 

Ah okay. Will test and post my results. Nice video!

[jnr4817]

I need to figure what to manage with process lasso, what to put to file and what to make high priority etc. the only thing on my pc is dcs and all its peripherals to okay the game in vr. 

[speed-of-heat]

For the 9800x3d largely leave most stuff alone, set dcs to high priority, have it avoid cpu 0 and 1 ( as they are the same physical core) as windows  processes tend to default there