investigating Aircraft coming close to player aircraft cause CPU spikes (leading to stutters) in VR since 2.9 - MT ( single + multiplayer)

[darkman222]

This started with 2.9 MT for me. Before it did not matter how close the other aircraft were. In my case its just fly bys. But its also when chasing aircraft which makes aiming difficult. Its in multiplayer only. In SP this issue is not happening. And its not only if the opponent is an F16. Its for all aircraft. So its not a LOD thing of the F16 specifically.

See the CPU graph of fpsVR. You see when the bandit comes close the CPU graph shows spikes that translate directly to the GPU, causing stutters.

It is on a multiplayer server, in the 1vs1 arena with only 2 aircraft in the area. Also it was morning, and just 3 or 4 players in total on the server that time.

I tried to disable Core Parking in the registry, that made it even worse, although I could not directly compare due to the nature of multiplayer sessions.

Seems to be a Multithread issue. See the same test with ST in the follow up post.

System is a :

-Intel 13900KS

-RTX 3090

-Varjo Aero VR headset

 

 

mobettameta's Dogfight Arena v1.76.1-20231030-112747.trk mobettameta's Dogfight Arena v1.76.1-20231031-094405.trk

Edited March 22, 2024 by darkman222

[darkman222]

Nobody else?
Might be a multi thread issue.

I did the same test in single thread. And no CPU spikes occurred. Besides that the CPU load in general is higher than in MT and thats why the GPU is struggling getting the images rendered in the given refresh rate.

If its a multithread issue some mod might want to move the thread there.

See the video with single threaded this time:

 

Edited November 2, 2023 by darkman222

[SloppyDog]

On 10/31/2023 at 2:37 PM, darkman222 said:

This started with 2.9 MT for me. Before it did not matter how close the other aircraft were. In my case its just fly bys. But its also when chasing aircraft which makes aiming difficult. Its in multiplayer only. In SP this issue is not happening. And its not only if the opponent is an F16. Its for all aircraft. So its not a LOD thing of the F16 specifically.

See the CPU graph of fpsVR. You see when the bandit comes close the CPU graph shows spikes that translate directly to the GPU, causing stutters.

It is on a multiplayer server, in the 1vs1 arena with only 2 aircraft in the area. Also it was morning, and just 3 or 4 players in total on the server that time.

I tried to disable Core Parking in the registry, that made it even worse, although I could not directly compare due to the nature of multiplayer sessions.

Seems to be a Multithread issue. See the same test with ST in the follow up post.

System is a :

-Intel 13900KS

-RTX 3090

-Varjo Aero VR headset

 

 

mobettameta's Dogfight Arena v1.76.1-20231030-112747.trk 14.87 MB · 3 downloads mobettameta's Dogfight Arena v1.76.1-20231031-094405.trk 21.5 MB · 3 downloads

 

I have a i7-4700K. 

I see that you have a "K" CPU as well. It means they have unlocked core frequencies, to allow overcloking.

In gaming, mine tries to go full throttle to better run the game, but enter in throttling since the CPU gets too hot, and the clock has to be throttled down in order to protect the CPU, but the CPU clock goes up again in order to better run the game, then goes down again, and keeps its cycle during gameplay. It results in stutters.

I don't know if it is possible in newer CPUs like yours, but what solved the problem for me was undervolting the CPU. This way you tell the CPU the maximum allowed power it can go, so it runs at the maximum allowed power, that you set, all the time during the game, with almost no spikes and stutters.

 I run in 2D, and yes, it is working better in Single Thread than in multi-thread for me as well.

 

[darkman222]

Thanks for you effort. But I seriously doubt that my PC coincidently and all of a sudden starts getting too hot when 2.9 was released the same time.

Today I played on  another server that had the best performance in the past showing the same behavior. Also the CPU spikes were gone when using ST. (But overall performance was of course worse than in MT)

DXdiag attached.

Can some Moderator please take over and move it at least to the MT bug section, as this clearly is a MT bug.

@BIGNEWY Thank you.

 

DxDiag.txt

Edited November 2, 2023 by darkman222

[FallingWithStyle]

Same here. I have tried everything that doesn’t involve changing Windows settings (not going to mess with the thread director or core unparking), but have given up for now. The only thing that made a noticeable difference was changing my RAM timings and speed to be slightly slower, and limiting the fps to 144.
 

Game still crashes randomly, usually when restarting or loading a level, and the stutters are very random but consistent in the big Syria cities with low flight. I think there’s a compounding issue with memory management and thread usage, it doesn’t feel like one issue.

Edited November 13, 2023 by FallingWithStyle

[BIGNEWY]

please try this and see if it helps 

 

[f14billy]

Another thing you can try is in the NVIDIA Settings, set pre rendered frames vr to 4. In my case it got rid of the stutters I was seeing when a nearby aircraft was deploying flares during a fly by at the merge.

Edited November 14, 2023 by f14billy

[T-Pickle]

I've noticed some spikes when "merging" with other aircraft, as if something is loading in and takes some time to do so. When staying in range and following them I couldn't reproduce the spikes reliably anymore, so aiming was usually fine. Sometimes it still felt like the spikes came back randomly, usually during heavy manoeuvring, possibly due to the rapid change of camera perspective and objects coming in and out of view. 

However, a similar thing happens when I'm close to scenery objects like buildings (the more the worse) and this time it won't simply go away, as if any building nearby has to load a lot of data again and again while flying past a city. The spikes can be anything between 5 and 20 fps loss, which is quite noticeable for me in VR with my FPS capped at 45. Graphics settings didn't seem to make a difference with the general impact (even anti-aliasing off), I use a moderate custom preset.

Game is on M.2 SSD, 32 GB of RAM, Intel i7-6700K, GTX 1080. Yes I know it is a bit older, but these issues seem to happen for people with much better systems as well so I just want to contribute my experience. Running MT and currently testing the new MT.lua @BIGNEWY has linked in another thread. So far no luck with the issues described above. Hope we can find a solution together at some point.

Edited November 15, 2023 by T-Pickle

[darkman222]

Quick heads up on the state of the issue.

With the new update DCS 2.9.1.48111 Open Beta either with or without MT.lua the issue persists. In the previous OB the MT.lua file even produced worse results. Now it does not matter any more.

Looks like our issue here is just the little brother of the issue other people have with massive CPU spikes all the time.

[darkman222]

DXDiag DCS Log from todays session. See the video in first post. Same issue all the time.

DxDiagDCS29 20231216.txt

dcs.log

Edited December 16, 2023 by darkman222

[Oilman100]

16 minutes ago, darkman222 said:

DXDiag DCS Log from todays session. See the video in first post. Same issue all the time.

DxDiagDCS29 20231216.txt 101.23 kB · 0 downloads

dcs.log 154.33 kB · 0 downloads

 

I have exactly the same issue as you and only in MT, ST runs perfect, I also get the spikes and notice that when the split second stutter occurs, Core 1 max’s out at 100%, then most lightly catches up and stutter ends and then repeats when it max’s out again. Have tried all the fixes suggested and still no joy.

[FallingWithStyle]

Just want to say this issue feels hugely better after returning to the game. Not sure what helped, but I think a big Nvidia driver update (huge version jump about a month ago) helped some issues alongside a new version of Intel management engine that went along with my BIOS update. Whatever it was, thank you @BIGNEWY

[darkman222]

@BIGNEWY No! The stutters with aircraft coming close to the player are still there as before. DCS 2.9.3.51704 - 22.02.2024 

The gazelle seems to have a good impact for testing. Another F16 does the same, but gives not that much of a CPU impact. As soon as I come close to the gazelle the CPU frame time goes up. Creating bad spikes in fpsVR which are visible as hefty stutters. Thats the yellow dashes in the GPU graph. Thats what looks like stutters in VR.

Especially in dogfighting it is really annoying when youre close to your opponent when smooth playback is needed you end up in a stutter mess.

See how the CPU frame time goes up every time

DxDiag2_9.txt stutters gazelle.trk

Edited March 19, 2024 by darkman222

[darkman222]

@BIGNEWY Can we please get a tag if it is being worked on or not? As soon as the ST exe will be taken away from us, we wont have any reference to test it with. It is definitely a MT issue here. Testing again with ST and no lags or spikes occurr on the CPU frame time graph. If there was MT optimisation done in the latest update it does not show up in improved texture and geometry streaming performance of displayed aircraft in MT.

I had a talk with another VR user who got hefty stutters in the DCS menu already. Switching to ST got rid of the stutters. But this is no permanent solution as ED is planning to take away the ST exe and the lower performance ST has compared to MT.

 

Edited March 19, 2024 by darkman222

[BIGNEWY]

I have marked the thread investigating. 

But when I try I don't see the same as you, I will ask other to try also. 

thank you 

[darkman222]

Thanks a lot.

Me running an Intel 13900KS with an RTX 4090. Vr Headset is a Varjo Aero. And second VR headset is the Pimax Crystal.

The user I talked to had a comparable system. Crystal, RTX 4090, 13900K ( not KS). He even had stutters in the menu which were gone using ST.

Another user on the dogfight server has the 7800x3d + 4090 + Reverb G2 without issues. Unfortunately he has a different headset and processor. I really am thinking about the intel processor at least contributing to it...

 

[BIGNEWY]

51 minutes ago, darkman222 said:

Thanks a lot.

Me running an Intel 13900KS with an RTX 4090. Vr Headset is a Varjo Aero. And second VR headset is the Pimax Crystal.

The user I talked to had a comparable system. Crystal, RTX 4090, 13900K ( not KS). He even had stutters in the menu which were gone using ST.

Another user on the dogfight server has the 7800x3d + 4090 + Reverb G2 without issues. Unfortunately he has a different headset and processor. I really am thinking about the intel processor at least contributing to it...

 

I have asked the MP team to check, I am running a RTX 3090 with an Intel® i9-10900K 

[darkman222]

Sorry for bumping it up again @BIGNEWY and @Lord Vader. But I did some more testing. I was suspecting the LOD switch factor to have effect on it as many signs point to the sutters coming from geometry and textures loading in and out of the CPU threads.

Different LOD switch factor settings produce different CPU frame time spikes and thus different stutters. Interestingly a factor of 1 which is default, produces the worst results. This was all done with the same the track replay.

In my previous test using the ST branch of DCS  DCS 2.9.3.51704 did not produce stutters as expected. We know that already.

But for this test I reverted to DCS 2.8.7.42718 OB where I remembered not to experience these stutters. And whatever was added to DCS from 2.8 MT to 2.9 MT contributed to this bad CPU frame time behavior a lot. I used the same mission as it was simple enough that I could run it in 2.8 too. So the results are pretty well comparable.

I am knowing that DCS is progessing. But from 2.8 MT to 2.9 MT something was introduced that made the LOD handling in MT worse. And the root cause seems not having been adressed since then.

Edit: My test reverting back to 2.8 at the present day is not irrelevant, as it clearly shows it's not a change in hardware or driver software that caused my well running system to perform worse with 2.9. because I get my performance back the very moment I revert to 2.8

 

 

Edited March 22, 2024 by darkman222

[BIGNEWY]

I will ask other testers to take a look again, however on my system I am unable to reproduce. 

Could I ask you to post your NVDIA control panel settings also 

We have a planned patch on the 3rd April please let us know if you see any difference once released

 

thank you 

[beacon]

On 11/15/2023 at 12:22 PM, T-Pickle said:

I've noticed some spikes when "merging" with other aircraft, as if something is loading in and takes some time to do so. When staying in range and following them I couldn't reproduce the spikes reliably anymore, so aiming was usually fine. Sometimes it still felt like the spikes came back randomly, usually during heavy manoeuvring, possibly due to the rapid change of camera perspective and objects coming in and out of view. 

However, a similar thing happens when I'm close to scenery objects like buildings (the more the worse) and this time it won't simply go away, as if any building nearby has to load a lot of data again and again while flying past a city. The spikes can be anything between 5 and 20 fps loss, which is quite noticeable for me in VR with my FPS capped at 45. Graphics settings didn't seem to make a difference with the general impact (even anti-aliasing off), I use a moderate custom preset.

Game is on M.2 SSD, 32 GB of RAM, Intel i7-6700K, GTX 1080. Yes I know it is a bit older, but these issues seem to happen for people with much better systems as well so I just want to contribute my experience. Running MT and currently testing the new MT.lua @BIGNEWY has linked in another thread. So far no luck with the issues described above. Hope we can find a solution together at some point.

 

hi there yes i have the same issue, but please upgrade to more ram. i had a flight in the weekend  an d my ram was 53 GB  i have 64GB . this was on 4ya mutliplayer server sirya map 

[darkman222]

I am aware of that textures loaded from the SDD into RAM also can lead to stutters. ( 64 GB memory are ordered btw) But in my testing, you can see in fpsVR that my memory is way below the limit and no factor.

@BIGNEWY@Lord VaderI waited a few updates now for a final test and some conclusions here. My test is like it was from the first time I started testing: Approaching a Gazelle over the ocean. (see the video above how I do it, you can also download a track file of it) I could have used another aircraft to fly close to, but the Gazelle seems to have the biggest impact on CPU frame time and texture loading and unloading.

Here are my inital tests about in which version the issue has emerged. I used the DCS updater to revert to older versions. So its all with the same hardware and drivers. Tested on the same day. No change in hardware or drivers affect that test this way.

Note: In fpsVR see the graph for individual CPUs, that with 2.9 DCS seems to start using the CPUs more evenly than it used to in 2.8. More about it towards the end of my test.

This leads to another test how to improve that behavior. A lower texture size of course helps people that are affected by it, but as long as the texture slider is for all textures, including cockpit textures, it does not help in the long run as long as we dont get separate sliders for cockpit and exterior textures.

I still wanted to share my test up to this point as this might give some further insight, without being a solution though.

 

 

 

Here comes my most important test. I was trying a method mentioned by Jabbers in his testing running DCS with arguments: --affinityF (4 cores) to --affinityFFF (12 cores)

That did not help in my case. It made DCS run unstable. Possibly because it disables random cores. Then I used process lasso and clicked "disable hyperthreading". That also disabled the efficency cores. Note, this only works with process lasso. Disabling the hyperthreading cores and efficiency cores in the BIOS also leads to an unstable running DCS and even the entire PC runs unstable. The only way to do this is with process lasso.

 

Now DCS runs only on these cores ( note the efficiency cores are also disabled) :

Now here are the results. The stuttering because of textures loading and unloading into the cores / threads is completely gone.

I tried ( multiple times, but without success) the MT.lua @BIGNEWYposted in the thread he started called: "Intel hybrid CPU Getting stutter? Please try this in multithreading DCS" but it did not help. Not sure if it was rather to fix core parking issues. Also it took me this long to figure that solution out because disabling cores via BIOS is no stable solution.

 

Sorry for the long post. But I felt that I needed to share these results in detail, as I spent quite some time gathering them.

 

I included a ZIP file of my system specs, a dxdiag and a dcslog with "allcores" and with "8cores" in the file name. So you have a file that was created using all cores and a file that was created when DCS just used 8 cores. As requested, I included my GeForce settings too. But I dont thing they matter and they are just the plain default anyway.

 

Also a track file is attached. I am not really sure if you can reproduce it on your systems as this behavior is barely visible with the DCS fps graph when trying this in 2D. Running DCS on a high resolution VR headset is of course demanding for the hardware and clearly shows if something is not running smoothly. I am willing to help with further testing for ED, trying LUA files or whatever needed. For now this is a satisfying solution as long as nothing changes and I can keep the performance using this core affinity with process lasso.

 

dxdiag_dcs_log__geforce_settings_hardware_darkman222.zip

stutter8cores.trk

Edited June 3, 2024 by darkman222

[KoN]

Can confirm if the F4 is around me I get lag in online servers . 

[darkman222]

The F4 is a different beast, concerning performance. This thread is only about stutters appearing to LOD switching, VRAM, caused by other jets coming close to the player.

[darkman222]

@BIGNEWYAny plans to limit the DCS.exe process only to the P Cores ( also exclude the hyperthreaded cores) for intel Hybrid CPUs for multithreaded DCS.exe ? After quite a good research on the Varjo forums it seems to help a lot of people especially people running high end VR headsets that show spikes in CPU frame time with hefty stutters.

I have been testing this for 2 weeks now on multiplayer servers that showed that behavior when getting close to other players. It had a great relief on my CPU frame times and no stutters, if it was caused by LOD switching related stutters.

 

Edit: Maybe a checkbox in the settings to bring DCS to P Cores only would be a helpful addition for users.

Also a separate toggle for exterior texture (jet) and interior( cockpit) resolution would be helpful to reduce stutters.

Edited June 14, 2024 by darkman222