VR Benchmarks (initial tests)

[Qcumber]

4 minutes ago, Mr_sukebe said:

Sorry, I don’t know enough about data exports to understand how you’d match up their precise timings with your render time info.

Regardless of that, it was very easy to see that maxing out any of those resources results in a significant degrade to fps.  The interesting bit is how to turn this understanding into a rough guideline on recommended settings for people with different rigs.

Yes. My main aim is to tease out the settings which impact most, irrespective of rig. It's not an easy task. Ideally it would involve using preset values and test tracks that people could run on their rigs and compare results. 

[Mr_sukebe]

Not easy.  I'm hoping that maybe we can bury that into here:

 

Instead of trying to play guesswork when someone opens a new thread to ask what has already been asked many times before, that we can direct them to a single post that provides key guidance, as against having to repeat the same process of Q&A.

 

[Qcumber]

4 minutes ago, Mr_sukebe said:

Not easy.  I'm hoping that maybe we can bury that into here:

 

Instead of trying to play guesswork when someone opens a new thread to ask what has already been asked many times before, that we can direct them to a single post that provides key guidance, as against having to repeat the same process of Q&A.

 

Yes. A single place with links to useful threads at the top. Updated regularly. 

[Qcumber]

 

My findings with various QVFR settings.

The test run is a flight over Cairo in an F-16 at about 300 ft. 

T2 (green) is without QVFR. Base resolution 2756 (meta app) and 1.2 in ODT (total resolution 3300 pixels). Mean GPU render time 11.5 ms.

T4 (light blue) is with QVFR centre x1.35, 0.22x0.22, periphery 0.37. Mean GPU render time 8.1 ms.

T7 (dark blue) is with QVFR centre x1.2, 0.8x0.8, periphery 0.3. Mean GPU render time 12.26 ms.

I had thought that T7 would be better than T2 but it is slightly worse. The only difference is that 64% of the resolution is at the equivalent of 3300 pixels. Whilst the remaining 36% is at 826 pixels. The total count should be ((3300 x 0.64) + (826 x 0.36)) = 2.4 Mp. Compared to 10.9 Mp without QVFR. There must be some overhead on GPU us when using QVFR.  

T4 has a total pixel count of ((2756 x 1.35) x 0.048) + ((2756 x 0.37) x 0.952) = 1.15 Mp. There is a very big increase in performance (mean 3.4 ms lower render time compared to T2). The best use of QVFR is to use a narrow foveated region. 

CPU render time is increased with QVFR as expected but this is relatively small.

Edited April 27 by Qcumber

[Mr_sukebe]

Nice set of tests.  Makes sense.  The T4 settings are very similar to my current setting.

[niru27]

On 4/21/2025 at 4:29 PM, Qcumber said:

That's a good point regarding CPU and GPU usage but I'm not sure how I can capture this alongside the other data. Any suggestions about software I could use? It would need to allow data export to a CSV file. 

It would be interesting to correlate usage with render times. My observation is that if GPU is maxed out then the render time exceeds 13.88ms and fps drops; which is exactly what you would expect. CPU usage would've harder to correlate due to multiple votes. 

Using Intel Present Mon you can log GPU Busy, GPU Utilization % and total Frame Time.

If Frame time > 13.89 and GPU Util > 100%, you're hitting GPU limits

If Frame time > 13.89 and GPU Util < 95%, you're hitting CPU limits

The problem in DCS is when there's a lot of units activity, it can spike CPU usage, so your frame time will go up. Perhaps we should tune settings to aim ~80% GPU usage so that we have headroom for sudden acitivities?

--------

Have you tested how much CPU overhead QV adds?

If you set peripheral to 1x and focus to 1x @ 0.01x0.01x, and compare that with QV disabled, that might tell us the minimum overhead required?

[Qcumber]

1 hour ago, niru27 said:

Using Intel Present Mon you can log GPU Busy, GPU Utilization % and total Frame Time.

If Frame time > 13.89 and GPU Util > 100%, you're hitting GPU limits

If Frame time > 13.89 and GPU Util < 95%, you're hitting CPU limits

The problem in DCS is when there's a lot of units activity, it can spike CPU usage, so your frame time will go up. Perhaps we should tune settings to aim ~80% GPU usage so that we have headroom for sudden acitivities?

--------

Have you tested how much CPU overhead QV adds?

If you set peripheral to 1x and focus to 1x @ 0.01x0.01x, and compare that with QV disabled, that might tell us the minimum overhead required?

 

Thanks. I'll try using Intel Present Mon. 

I did try running QVFR at x1 and x1. I'll try and dig out the data. It was part of another test so it would not necessarily compare to this one. 

I really need to try running these tests again when I can also see CPU and GPU usage. 

[Qcumber]

1 hour ago, niru27 said:

The problem in DCS is when there's a lot of units activity, it can spike CPU usage, so your frame time will go up. Perhaps we should tune settings to aim ~80% GPU usage so that we have headroom for sudden acitivities?

Yes. My setup can handle most things in DCS but when there are too many units etc the CPU render time starts to spike and that then impacts on GPU latency. It's a real mess. The problem is it can be hard to find a track which allows you to look at this objectively. I have done some work on this but have not yet posted it as it looks a bit messy. I'll try to do this later. 

[niru27]

You can use this mission to reliably get CPU limited right at the (hot) start of the first mission, with a lot of activity. Was getting ~50fps with 12400 and went to 220+ with 9950x3D

I always get GPU bound at the beginning of Raven One mission 2 "Riders on the Storm"

But, with lower graphics, CPU gets additional headroom, and vice versa, so it is just a balancing act between CPU frame time and GPU frame time

Edited April 27 by niru27

[Qcumber]

This is a test I did to compare different CPU loads. It is based on a track flying parallel to Utah beach in the instant action mission for the P-47 free flight on D-Day. Out to sea are a lot of units and, combined with explosions and smoke on the beach, creates an intense situation. However, flying a mile or so inland is much less demanding. Theses are three recordings based on looking looking right towards farmland (blue, T2), left towards the beach (green, T6), or looking straight (purple, T10). 

Looking towards the farmland (T2) is easy on the system as CPU and GPU load are low. so this has the lowest latencies. 

Looking straight (T10) is mixed but overall 72 fps is easily maintained. 

Looking towards Utah beach (T6) creates big problems for the CPU with frequent spikes and very variable render times.

Overall, FPS is maintained at 72 fps but micro stutter was prevalent in T6. There was none in T2. 

 

 

Edited April 27 by Qcumber

[actually_fred]

New version with bugfixes: https://github.com/fredemmott/XRFrameTools/releases/tag/v0.2.0

Also worth noting that https://github.com/fredemmott/XRFrameTools?tab=readme-ov-file#why-should-i-use-this-instead-of-my-favorite-tool-for-non-vr-games applies to the several non-VR tools mentioned in this thread.

[scommander2]

Few more questions if they are applied to DCS:

1.  Is the test based on the DCS default DLSS preset model, or use Preset J/K?  

2. Reflex/Frame Generation help?

[Qcumber]

7 hours ago, scommander2 said:

Few more questions if they are applied to DCS:

1.  Is the test based on the DCS default DLSS preset model, or use Preset J/K?  

2. Reflex/Frame Generation help?

1) The tests above should state if DLSS is used and which preset. My default setup now is to use DLSS quality with preset K. I am in the process of comparing different presets which l will post when completed. There is a significant difference between "C" and "K". 

 

2) If you are referring to reproduction then no, this is not applied. In terms of DLSS then frame generation does not work with VR. 

[scommander2]

Thanks for the comments and good to learn more!!

[ChariotOfFLAME]

@Qcumber you may have already posted this, but do you have a settings table with avg % impact of each setting higher than lowest? I feel like you’ve got enough data to do that at this point? I also understand how much work creating that can be

[Qcumber]

36 minutes ago, ChariotOfFLAME said:

@Qcumber you may have already posted this, but do you have a settings table with avg % impact of each setting higher than lowest? I feel like you’ve got enough data to do that at this point? I also understand how much work creating that can be

Not yet but I intend to. I've not had much time to work in this recently but hopefully I can get something together soon. 

[Hempstead]

@Qcumber Thanks for testing & publishing these. My new gaming rig is very similar to yours, except a slightly higher spec. that really doesn't matter to DCS. Saves me a lot of time tuning this stuff!