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

[DonMuskopf 32]

1 hour ago, Ray Proudfoot said:

I’m surprised you’ve set a limiter of 120

Using unlimited in sim for some reason disrupts G-Sync when using NCP's frame rate limiter. Rapid changes in viewpoint result in wavy scenery textures and makes G-Sync much less fluid. The idea of using 120 vs unlimited wasn't my original idea. I believe it came up in a forum discussion with a developer (can't remember who).  At some point in time it became apparent that P3D's internal limiter was not completely effective, which  contributed to 100% utilization on LP0. Setting the limit to 120 and using NCP at 35 (or 30) fixed that problem. It's reliably worked for me probably since V4.5. It wasn't until v5.2 that the 100% utilization problem surfaced again.

To be honest I haven't really noticed any real difference between 120 and unlimited as far as performance goes. As long as LP0 is running less than 100%, I get good smooth performance.

 

 

[gianluk81 3]

How strange ... I, like others, have the problem of core 0 at 100% on i9 9900kf with ht off. Today I installed the fslabs series and ... I launch the sim with the 320fsl and the use of core 0 is around 40/50% constant. If I use a pmdg or the maddogX, but also the aircraft by default, the use of core 0 is, like al
usually, 100% constant. I don't understand why, with fslabs core 0 goes down to 50%.

[CaptKornDog 284]

3 hours ago, DonMuskopf said:

Using unlimited in sim for some reason disrupts G-Sync when using NCP's frame rate limiter. Rapid changes in viewpoint result in wavy scenery textures and makes G-Sync much less fluid. The idea of using 120 vs unlimited wasn't my original idea. I believe it came up in a forum discussion with a developer (can't remember who).  At some point in time it became apparent that P3D's internal limiter was not completely effective, which  contributed to 100% utilization on LP0. Setting the limit to 120 and using NCP at 35 (or 30) fixed that problem. It's reliably worked for me probably since V4.5. It wasn't until v5.2 that the 100% utilization problem surfaced again.

To be honest I haven't really noticed any real difference between 120 and unlimited as far as performance goes. As long as LP0 is running less than 100%, I get good smooth performance.

 

 

Same here, setting to unlimited seems to disable/ignore G-Sync for me.  Can't explain it or figure out why.

Edited June 20, 2021 by CaptKornDog

[Dreamflight767 312]

9 hours ago, Ray Proudfoot said:

@Dreamflight767, in my case the PC was delivered with HT enabled.

If it’s enabled then for each physical core there are two virtual ones.

You enter a binary set of numbers into a calculator and the Decimal value shown is the value you enter into P3D.cfg in AffinityMask.

So, for example I have a six core i7-8086K. I want only one virtual processor to be enabled for each physical one. To calculate this use binary where 0=off and 1=on.

01 01 01 01 01 01

Open the Windows calculator in Programmer mode. Choose the Bin option. Enter the values above and press Decimal. It will show 1365. That’s the value you enter in Prepar3d.cfg

[JOBSCHEDULER]

AffinityMask=1365

But if HT is off and you still want to enter a value type 111111 and the Decimal value is 63.

For an eight core with HT off it would be 11111111 which equates to 255. With HT on it would be 0101010101010101 = 21845.

Hope that helps.

Thank you for taking the time to type this out.  I have an i77700K.  I didn't realize there was two virtual cores, I only thought there was one.  So when I have HT enable, I get 8 "cores".

Guess I'll have to play around.

[787flyer 633]

 

2 hours ago, Dreamflight767 said:

Thank you for taking the time to type this out.  I have an i77700K.  I didn't realize there was two virtual cores, I only thought there was one.  So when I have HT enable, I get 8 "cores".

Guess I'll have to play around.

Hi Aaron,

Just a quick pointer that what Ray posted in his last example was referring to a true 8 core CPU such as an i9-11900 CPU or similar where you would have 8 physical cores and 16 (LPs) logical processors which means its capable of running 16 threads.

 

Wrt to your i7-7700K, that has 4 cores and 8 LPs (also 8 threads when HT is enabled)

With HT enabled on this CPU, this would create an Affinity Mask of 85  that represents 01 01 01 01. With HT disabled your AM would be 15 i.e. 1 1 1 1

If in any doubt as to what AM to set, you can calculate this via the Windows calculator as Ray recommended or alternatively refer to this small freeware web tool here that does conversion between AM values and CPU allocation in both directions. : http://store.fsxtimes.com/cal-cam.php

 

 

 

[Ray Proudfoot 8,958]

7 hours ago, DonMuskopf said:

Using unlimited in sim for some reason disrupts G-Sync when using NCP's frame rate limiter.

I suspected it might be related to G-Sync which is something I’ve never used. Experimentation seems to be the key especially with 5.2.

[Ray Proudfoot 8,958]

3 hours ago, Dreamflight767 said:

Thank you for taking the time to type this out.  I have an i77700K.  I didn't realize there was two virtual cores, I only thought there was one.  So when I have HT enable, I get 8 "cores".

Guess I'll have to play around.

You’re welcome. It’s very easy to check the results of any AM setting. Display Task Manager - Performance tab and you should see the chosen LPs showing activity with the others just ticking along close to zero.

[GSalden 4,657]

Ray, it is good to see that your system is now running better, but the main issue as the title says, that the CPU is more taxed in v5.2 and the GPU less in v5.2 compared to v5.1 remains.

If I have read it correct you simply had HT enabled without a correct AM.

Others, like me, with a good working AM are still seeing the issue.

So writing that that issue is Solved might be a little premature…

 

[xTiMm_ 50]

Can anybody help me out with AM for my i9 9900k running with no HT (works smoother!)

 

I see a way bigger load on CPU0 (or 1) than on the others. I don´t have stutters but sometimes in heavy regions, Autogen takes ages to load in.

Thx 🙂 

[GSalden 4,657]

22 minutes ago, xTiMm_ said:

Can anybody help me out with AM for my i9 9900k running with no HT (works smoother!)

 

I see a way bigger load on CPU0 (or 1) than on the others. I don´t have stutters but sometimes in heavy regions, Autogen takes ages to load in.

Thx 🙂 

This looks good.

Your main thread core will always be taxed more than the others. 
As long as it does not stay at 100% then you will have headroom and no stutters.

Heavy regions have more objects, more AG, detailed PR scenery. If you set your sliders so that in the heaviest situation you do not have 100% continues load then you are good to go.

 

[Afterburner 374]

27 minutes ago, xTiMm_ said:

Can anybody help me out with AM for my i9 9900k running with no HT (works smoother!)

Are you currently using any AM settings?

If you experience slow autogen loading in dense areas, I would turn off AM completely and also activate hyperthreading. Playing around with AM may initially give you some FPS boost, but it comes at a cost that is not instantly visible.

[Ray Proudfoot 8,958]

1 hour ago, GSalden said:

If I have read it correct you simply had HT enabled without a correct AM.

Others, like me, with a good working AM are still seeing the issue.

So writing that that issue is Solved might be a little premature…

The [JOBSCHEDULER] entry had a space in it so the AM setting was ignored.

It's solved for me. Others have never had the problem. I can only speak for myself and what has worked for me.

It's possible your multi-monitor setup could be an issue but @Pete Dowson also has that and has no issues. It also looks like your running two PCs for P3D. Again, an unusual setup.

Maybe wait and see what the 5.2 hotifx brings?

[GSalden 4,657]

Multiple users in this thread and at the LM thread have reported the more CPU load / less GPU load issue. 
And I am one of them as I showed with a vid.

According to Marcus the HF is solving this so load will be back to v5.1 level.

Myself I prefer they take it one step further and take more load of the main thread core and put it on the GPU. The 30 series are capable of so much more than what they need to do now. Even with 2x 4K as one wide view and all GPU taxing settings at max.

Hopefully LM will release a HF that all of us pleases ✈️

[micstatic 4,319]

4 minutes ago, GSalden said:

Multiple users in this thread and at the LM thread have reported the more CPU load / less GPU load issue. 
And I am one of them as I showed with a vid.

According to Marcus the HF is solving this so load will be back to v5.1 level.

Myself I prefer they take it one step further and take more load of the main thread core and put it on the GPU. The 30 series are capable of so much more than what they need to do now. Even with 2x 4K as one wide view and all GPU taxing settings at max.

Hopefully LM will release a HF that all of us pleases ✈️

I’m also massively in favor of this 

[mazelda 124]

Sorry to jump in on this topic, but I find it very interesting but some of it is a bit above my head.

Is there somewhere I can read up on affinity masks and Other performance tweaks specifically for P3d.

I have a couple of specific questions at the moment that maybe someone can answer while i try reading up 

1. I have just moved over to the dark side and i'm now on a Ryzen 5950x, Can you turn hyperthreading off on a Ryzen? I knew how to do it on an intel bios but I think AMD call it something different?

2.When I select task manager and select performance I only get overall cpu utilisation, how do i bring up  individual cpu cores/threads?

Ignore this one I have figured it out🙄

3. Specifically for the Ryzen cpu's is there anything i should be looking at that helps get the best out of P3d

4. What would be the best things to start reading up on?

I should add that I'm currently running 5.1 with loads of ai via aig and use ai traffic controller to have the number set to a reasonable amount, I have not added any tweaks or shaders and I'm very happy with the performance  so far.

But its always good looking for that little bit more.

I am also running a 3090 with 32gb  3600 ram and docp enabled

Again sorry for jumping on this but I thought it might be better here than the hardware/software forum as this may have some good benefits for me and the thread is very active with some very knowledgeable  people posting 

 

regards

Edited June 20, 2021 by mazelda