Intel hybrid CPU Getting stutter? Please try this in multithreading DCS

[BIGNEWY]

Hi all, 

we have been looking into the stutter issue some users are seeing, as you may be aware unparking CPU cores has helped many but this is not a ideal solution for end users. 

Please try the lua file below and give us some feedback. 

mt.lua

place in saved games\DCS\Config folder

this should unbind the render thread from the highest performing core. 

Based on the feedback we get we could make this default for some CPU's 

thank you

[Hiob]

I recently experienced some issues that may or may not be related to this. 

I'm running a 5900X - so no e-/p-core stuff going on, but I sometimes have regular frametime spikes (cpu-bound) on a solo flight. Single player, 2D, MT, no other aircraft in sight. 

The frametime diagramm looks like a elektro cardio diagramm with a massive ft-spike every second or so. The average fps is still ~120, but the 1% lows are 10 fps (ten). Needless to say, that makes a pretty stuttery experience, regardless of the average FPS.

This issue can sometimes be solved by waiting (after a couple of minutes it sorts itself out) or by rebooting.

I wonder if that is related (cpu-scheduling?) or something completely different.

Edit: of course, I'm going to try the above .lua - even if it isn't specificly tailored to my problem.

Edited November 14, 2023 by Hiob

[MoleUK]

21 minutes ago, Hiob said:

I recently experienced some issues that may or may not be related to this. 

I'm running a 5900X - so no e-/p-core stuff going on, but I sometimes have regular frametime spikes (cpu-bound) on a solo flight. Single player, 2D, MT, no other aircraft in sight. 

The frametime diagramm looks like a elektro cardio diagramm with a massive ft-spike every second or so. The average fps is still ~120, but the 1% lows are 10 fps (ten). Needless to say, that makes a pretty stuttery experience, regardless of the average FPS.

This issue can sometimes be solved by waiting (after a couple of minutes it sorts itself out) or by rebooting.

I wonder if that is related (cpu-scheduling?) or something completely different.

Edit: of course, I'm going to try the above .lua - even if it isn't specificly tailored to my problem.

 

You tried locking DCS to only running on one CCD? Might be worth a shot as well.

Edited November 14, 2023 by MoleUK

[Bouwplan]

@BIGNEWY Intel 12th gen and up require Windows 11 to unpark their cores. Have recently 'solved' this issue by myself and with a bit of research found the answer.

I did try to unpark cores in windows 10 but for newer Intel this does not work as windows 10 can't handle the P/E core stuff and even by trying to force the cores to unpark, it simply will not.

This is where Windows 11 comes in as it is more optimized for the P/E cores and will have them all unparked and fix the issue for those that are on windows 10 or lower.

Edited November 14, 2023 by Bouwplan

[Lurker]

5 minutes ago, Bouwplan said:

@BIGNEWY Intel 12th gen and up require Windows 11 to unpark their cores. Have recently 'solved' this issue by myself and with a bit of research found the answer.

I did try to unpark cores in windows 10 but for newer Intel this does not work as windows 10 can't handle the P/E core stuff and even by trying to force the cores to unpark, it simply will not.

This is where Windows 11 comes in as it is more optimized for the P/E cores and will have them all unparked and fix the issue for those that are on windows 10 or lower.

 

This should NOT be the case. The thread scheduler for Windows 11 exists and should fully work in Windows 10 as well. In fact this was patched in by Microsoft soon after the 12th gen Intel CPUs were released. 

[Bouwplan]

12 minutes ago, Lurker said:

This should NOT be the case. The thread scheduler for Windows 11 exists and should fully work in Windows 10 as well. In fact this was patched in by Microsoft soon after the 12th gen Intel CPUs were released. 

Yeah, so yell at Microsoft, but you need win11 these days. Its hardwired in Win10. I already tried all the options that were out there to unpark my 13700k (RegEdit, power management and software).

[Outlaw]

@BouwplanUse this if no other solution worked. License is very cheap and works with my 13900k 

 

https://bitsum.com/parkcontrol/

[NIGHTHAWK1]

Tried the MT Lua.  No joy.  Fact is I lost dcs.logthe new UI for MT.  cant set DLSS etc.   Win 10, I9-12900k

Please Disregard -- I forgot I went back to 2.8.7 .   Senior Moment

Edited November 14, 2023 by NIGHTHAWK1

[Bouwplan]

7 minutes ago, Outlaw said:

@BouwplanUse this if no other solution worked. License is very cheap and works with my 13900k 

 

https://bitsum.com/parkcontrol/

Thanks, but the windows 11 upgrade was free and has been done a while ago. And as i said, i already tried using software.

[raven06a]

Just tried the Lua file and can confirm that the problem was solved for me (intel 12700). No constant "saw teeth" stutters like the ones I was getting before trying the file. Only stutters I'm getting are the expectable ones on a maxed out 3070. 

I'll report after further use if the problem stays solved. 

Thanks!

Edited November 14, 2023 by raven06a

[BIGNEWY]

4 minutes ago, raven06a said:

Just tried the Lua file and can confirm that the problem was solved for me (intel 12700). No constant "saw teeth" stutters like the ones I was getting before trying the file. Only stutters I'm getting are the expectable ones on a maxed out 3070. 

I'll report after further use if the problem stays solved. 

Thanks!

 

thank you for the quick feedback. 

[YoYo]

Now I have forced all unparked cores (and the issue isnt present), will back my settings to default and I’ll check this Lua file. Thanks for your work!

[Buzz313th]

Just tried the Mt.LUA file and it caused an increase in stuttering.

Details below

12900k, RTX4080, 64gigs, Quest3 @ 4800x2600 in link app

No Mods

MT 2.9

No OpenXR apps or hacks

As VR vanilla as can get

 

My normal method for launching DCS 2.9 MT, which gives me the best experience is to simply launch DCS and as soon as the app loads and before I put on my HMD, I simply open task manager and disable core #8 for DCS in Affinity mask.  I simply leave it disabled for the rest of the session.

When I tried the MT.Lua file, I placed the file in to SavedGames/DCS/Config and launched DCS as I normally do, but without disabling core #8 to see if the MT.Lua file fixes the need to disable that core.   Upon putting my HMD on the menus looked normal.  I loaded a simple mission in Nevada with an F16C airstart over Creech looking towards the Vegas area.  When the mission loaded and I was sitting in the pit, I got a big long pause with a rotating hourglass.  This is typically what I see if I do not disable core#8.  So I alt tabbed out of DCS, opened task manager and disabled core#8.  Went back in to DCS and did some simple aileron rolls while looking out the side of the canopy, which method I use to show the highest degree of stuttering as the Nevada landscape scrolls around my view.  I noticed a definitive increase in stuttering both consistently and intensively.   I closed DCS, removed the MT.Lua file from savedgames/dcs/config and restarted DCS, disabled core#8 and loaded the same mission to confirm that performance should be better.   Upon mission load, I noticed that I had no Hotas control on my throttle.  Took HMD off and noticed I had no status lights on my throttle.  Restarted throttle by Powe cycling my throttle via unplugging and re-plugging the USB cable.  Went back in game and confirmed that performance and stuttering was similar to baseline before trying MT.Lua.

I have never lost my Hotas before, especially without restarting my computer.  I assume this was simply a coincidence and not related to the use of MT.Lua.   

 

If ED could use more info from me such as logs, or track files then just reach out, I am here to help if I can.

 

Cheers

Edited November 14, 2023 by Buzz313th

[Bazz_Mulder]

I've tried this hotfix even that I'm on AMD 5800x3d (64gb+4090, Win11), and I see positive results.

I've tested it in the same environment and with this file I see that not only the first core is loaded at 100% which was the case before, but load is spreed across other cores. Also, there is less frame time fluctuation and slightly more fps.

Edited November 14, 2023 by Bazz_Mulder

[Thor_HUN]

Unfortunately, it's worse than before.

Test video

[BIGNEWY]

We appreciate the feedback, keep it coming and let us know what cpu you are using. 

thank you

[TonyG]

Is it worth including which version of windows as well?  10/11?

[Red_Pilot]

dcs.logdcs.log.old

Unfortunately, this did not help me. I still have serious problems at least on the Normandy map

i7-12700KF / 3070ti / 32 g RAM / W10

Edited November 14, 2023 by Red_Pilot

[paul2437]

Hi

No change for me Pimax Crystal RTX 4090 i9 13900KF running @ 5.60 64GB Ram running @5200 micro stutters in 2d and VR. After removing the new LUA no difference. 

 

[YoYo]

8th and 10th core 100% (13900K), rest was sleeping however Id like to have to test more longer and during the play.

I would also like to remind everyone that this problem did not always appear. Sometimes it's ok, sometimes it's not, so after one run of DCS you can't simply conclude that everything is ok and problem gone. Test this new lua for a few days, if it doesn't appear, maybe it's ok, if the problem appears, it's not a solution.

Edited November 14, 2023 by YoYo

[Kazputin]

4 hours ago, Hiob said:

I recently experienced some issues that may or may not be related to this. 

I'm running a 5900X - so no e-/p-core stuff going on, but I sometimes have regular frametime spikes (cpu-bound) on a solo flight. Single player, 2D, MT, no other aircraft in sight. 

The frametime diagramm looks like a elektro cardio diagramm with a massive ft-spike every second or so. The average fps is still ~120, but the 1% lows are 10 fps (ten). Needless to say, that makes a pretty stuttery experience, regardless of the average FPS.

This issue can sometimes be solved by waiting (after a couple of minutes it sorts itself out) or by rebooting.

I wonder if that is related (cpu-scheduling?) or something completely different.

Edit: of course, I'm going to try the above .lua - even if it isn't specificly tailored to my problem.

 

Also on a 5900x and see similar issues even in single player, also strange graphical artifacts like it goes overcast or one of the new tints gets applied for a while. I will try the mt.lua thing later to see how that goes.

[varyag087]

Noticable improvement on my i5-9500F, thanks for the .lua

[Red_Pilot]

And what is very strange is that I continue to have very high power consumption in MT when I just open the F10 map, no matter on Normandy or on PG.

[Buzz313th]

Question to @BIGNEWY.

Considering that some users have reported inconsistent performance results across different sessions under the same settings.  Is it possible that the team is using some sort of AI learning to manage the CPU cores and other system resources for best performance?

[Hiob]

5 hours ago, Hiob said:

I recently experienced some issues that may or may not be related to this. 

I'm running a 5900X - so no e-/p-core stuff going on, but I sometimes have regular frametime spikes (cpu-bound) on a solo flight. Single player, 2D, MT, no other aircraft in sight. 

The frametime diagramm looks like a elektro cardio diagramm with a massive ft-spike every second or so. The average fps is still ~120, but the 1% lows are 10 fps (ten). Needless to say, that makes a pretty stuttery experience, regardless of the average FPS.

This issue can sometimes be solved by waiting (after a couple of minutes it sorts itself out) or by rebooting.

I wonder if that is related (cpu-scheduling?) or something completely different.

Edit: of course, I'm going to try the above .lua - even if it isn't specificly tailored to my problem.

 

Ok, I tried the .lua. Since my issue was on/off and not persistent, I can't really tell if it is gone for good. From initial testing, the ft-spikes didn't show up. I don't see much of a difference though. Load spread across the cores seems to be similar to before. If anything the 1%-lows might me a bit better than before. But that could very well be just confirmation bias.

Nope, sorry - scrap that. The spikes returned the same way as before. No difference. On the third try I had the stutters again with the same 10 fps lows (=100ms ft-spikes). 5900X.

Edited November 14, 2023 by Hiob