CPU core 0 used more than GPU in v5.2.

[GSalden 4,657]

33 minutes ago, Ray Proudfoot said:

@mpo910, Marcus. I've posted my findings on LM forum in this topic. What I think is significant is the problem also occurs with default Ai so LM should be able to replicate the problem.

If they fix it for default Ai then AIG stuff should also benefit.

@GSalden, so many abbreviations. PSXT and RT???

@Ray Proudfoot

PSXseeconTraffic and RealTraffic = Live AI traffic

https://www.lekseecon.nl

 

[TextRich 88]

2 hours ago, craigeaglefire said:

out of interest Ray @Ray Proudfoot, did you do a complete uninstall of 5.1 & complete 5.2 install (using provided complete full installer around 20GB size)?
 
my 9700K is definitely not getting hammered on any CPU core - & using plenty of addons & using AI traffic,

in fact, the performance is really good with 5.2 (& using 2080 Ti with 64GB RAM…

This is just my perspective, but I am a brand new P3D v5 user.  I have installed 5.2 and nothing prior to it.  I am getting the same 100% CPU0 issue.  So I don't think it is a result of an unsatisfactory upgrading process from 5.1 to 5.2.

@mpo910 has a good point in that the results looked good to LM with a vanilla install, and @GSalden has a good point that the vast majority of us has at least one add-on.  This got me to thinking: what add-on is so must-have that the vast majority of us use it?  FSUIPC 6.

So I just did my own testing using a FSUIPC 6 autosaved flight that had me on approach to a coastal airport.  I tested it as normal to ensure I got stutters every minute or so.  Referring to that save:

-Setting my graphics to their lowest possible did not help the CPU0 100% issue.  It did make my sim look like MSFS 98 LOL.

-I did not run any third-party applications that requires a running program (for example, Active Sky).

-I unchecked all the Add-Ons in P3D and made the required sim restart.  I then enabled FSUIPC 6 only.

-I hid CTRL+Z, set frames per second to 120 maximum (I don't get more than 80 but just in case Unlimited makes it wonky), and am using the Nvidia Studio Driver (not Game Driver).

What I have just found:

-FSUIPC 6's Auto Save feature (available only in the paid version) stutters the sim for a second, every time.  I had it set to every 60 seconds.  I changed it to every 5 seconds and yup, the sim stuttered every 5 seconds, and all the save files had my POV at the instant the sim stuttered.  When I disabled it, the stutters went away.  I'll post about this in the FSUIPC support form in case it is a known issue on their end.

-CPU0 doesn't hit 100% until when the sim loading progress bar is at around 50%-60% - "Loading Terrain Data."  The "Loading Scenery Objects" before that do not affect it.  "Auto Generating Memory," "Loading Air and Sea Traffic" and "Compiling.." (too fast for me to read it) do not affect CPU0.  So traffic does NOT cause the CPU0 issue, since CPU0 hit over 100% way before traffic was loaded.  Whatever got loaded in the middle of Loading Terrain Data causes it.  Hopefully LM has isolated the cause for its hotfix.

Other than that stutter every minute, I have not really had performance concerns other than that 100% CPU0 and my 2080Ti not having more than 3.5 GB VRAM being used.

EDIT: I just realized I should have tested this with all settings to their lowest.  I just did that and it was terrible, stuttering every two seconds.  I guess for my machine it has to be run with much higher settings.  I also tried disabling all my add-on scenery (mostly Orbx).  Nope, still CPU0 at 100%.  I tried a default aircraft.  Same issue.

Edited June 17, 2021 by TextRich
Additional testing.

[krich 8]

If this is linked to AI, that is interesting because I have all traffic turned off and don't have any AI (such as AIG) currently installed.

[TextRich 88]

15 minutes ago, krich said:

If this is linked to AI, that is interesting because I have all traffic turned off and don't have any AI (such as AIG) currently installed.

In my post above yours, I don't believe it has anything to do with any AI traffic.

[Pete Dowson 558]

1 hour ago, TextRich said:

-FSUIPC 6's Auto Save feature (available only in the paid version) stutters the sim for a second, every time.  I had it set to every 60 seconds.  I changed it to every 5 seconds and yup, the sim stuttered every 5 seconds, and all the save files had my POV at the instant the sim stuttered.  When I disabled it, the stutters went away.  I'll post about this in the FSUIPC support form in case it is a known issue on their end.

This is a well know result of the saving of the files for the flight. FSUIPC does nothing except send the save request to P3D and delete the oldest in the cycle.

The same would occur if you did it manually (plus the time to enter the menus etc) -- the extra in FSUIPC is just the deletion of the oldest save in the cycle. It has always been the case and the code in FSUIPC can't be any more efficient as it is so simple. You could stop the deletion of the oldest in cycle if you want -- just use the "also save" option instead of the regular cycle of a selected number.

Pete

 

[Pete Dowson 558]

41 minutes ago, TextRich said:

In my post above yours, I don't believe it has anything to do with any AI traffic.

Well, Ray Proudfoot's testing shows a vast difference with and without AI traffic. The quantity of AI traffic appears irrelevant though.

Pete

 

[Kaiii3 1,979]

short update from my side (for those not on the AIG Forum)

 

During testing I had no performance impact (1/3 of my AI is already SimConnect based), and so far no reports from internal members as well. I can not trest it currently myself, but I will look into the reports I get and see if i cna help LM find a solution.

So far it seems only BGL based traffic is causing this, correct?

 

[SteveW 1,792]

3 hours ago, Ray Proudfoot said:

I've posted my findings on LM forum in this topic.

Ray, in your screenshots on that thread, Logical Processor 0 and LP1 are both nearly fully utilised. Since that is an Hyperthreading CPU, and LP0 and LP1 share core 0, these are competing for throughput on core 0. If you move the throughput of LP1 from core 0 to another core the main task on LP0 will get full throughput from core 0. You said in an earlier post you were using an Affinity Mask through SIMStarter to assign the AM. I would check that your AM assigns only one LP per core.

[TextRich 88]

1 hour ago, Pete Dowson said:

This is a well know result of the saving of the files for the flight. FSUIPC does nothing except send the save request to P3D and delete the oldest in the cycle.

The same would occur if you did it manually (plus the time to enter the menus etc) -- the extra in FSUIPC is just the deletion of the oldest save in the cycle. It has always been the case and the code in FSUIPC can't be any more efficient as it is so simple. You could stop the deletion of the oldest in cycle if you want -- just use the "also save" option instead of the regular cycle of a selected number.

 

Thank you for getting back to me.  I just racked my brain for a few minutes wondering why AutoSaving was so smooth in FSX but not P3D and I realized something: I have not yet told my antivirus to exclude Lockheed Martin folders like I did with FSX folders.  So I have just done that.  I just tested again and the now stuttering is much more brief when AutoSave is being used, and Also Save has even shorter stuttering and sometimes it does not stutter.  Have I overlooked something else?  Perhaps P3D 5.2 just isn't as efficient with saving as FSX was for me.  All in all, it is definitely an improvement and I'm glad as I need it since I tend to forget to save at critical moments before a sim crash.

Edited June 17, 2021 by TextRich
I forgot something.

[chapstick 2,757]

I'm not convinced there's a universal problem with 5.2 here. I'm not seeing a difference in CPU usage compared to 5.1 that makes me think something is wrong--I'm getting higher FPS and lower VRAM usage. I say that not to be one of those "not happening to me so everything is fine" people but it doesn't seem like such a clear-cut issue that LM is going reproduce and fix. 

[TextRich 88]

7 minutes ago, Chapstick said:

I'm not convinced there's a universal problem with 5.2 here. I'm not seeing a difference in CPU usage compared to 5.1 that makes me think something is wrong--I'm getting higher FPS and lower VRAM usage. I say that not to be one of those "not happening to me so everything is fine" people but it doesn't seem like such a clear-cut issue that LM is going reproduce and fix. 

I think for some of the posters here the concern is that the CPU workload isn't efficient (it is actually risky for the hardware) and that graphic cards aren't picking up as much workload as they used to in 5.1.  I personally just started with 5.2 but that is my understanding so far from 5.1 to 5.2 upgraders.  Other than the FSUIPC stuttering @Pete Dowson just assisted me with, 5.2 has been amazing for me and I am extremely happy with my upgrade from FSX:SE.

[Ray Proudfoot 8,956]

@SteveW, after discussing this a couple of days ago I did what Dirk98 suggested and put it in the [Job Scheduler] section of P3D.cfg rather than in SimStarterNG.

It's currently 1365 which equates to...

010101010101 for a 6 core 12 logical processor CPU. i7-8086K running at 4.6Ghz.

I'm happy to change it for whatever you suggest.

[Ray Proudfoot 8,956]

14 minutes ago, Chapstick said:

I'm not convinced there's a universal problem with 5.2 here

I'm not making any judgements about 5.2 for everyone. Just reporting what I'm seeing by testing different scenarios. I've proved beyond reasonable doubt that on my machine disabling all Ai BGLs eliminates the problem. But if I enable even the default Ai BGL the problem resurfaces.

@Kaiii3, I have some Ai injected via Traffic Controller but not much. 98% is BGL-based.

[Kaiii3 1,979]

some other Idea:

Start P3D

Open Task Manager -> Details and set the Cores for the Prepar3d.exe manually to what you need.

[Pete Dowson 558]

36 minutes ago, Kaiii3 said:

During testing I had no performance impact (1/3 of my AI is already SimConnect based), and so far no reports from internal members as well. I can not trest it currently myself, but I will look into the reports I get and see if i cna help LM find a solution.

So far it seems only BGL based traffic is causing this, correct?

This is shown by Ray's testing, with a Carenado add-on aircraft, but so far I see no problems in 5.2, using AIG traffic to its full. in fact I am one of those who are so far quite pleased with 5.2 -- it seems to be smoother for me. However, this is on very limited experience with it so far (I'm rather distracted too much by the Euros football competition)

Pete

 

Edited June 17, 2021 by Pete Dowson