P3D V5.4 High CPU Usage & Temps

[dave2013 1,158]

Personally, I wouldn't go fiddling with CPU voltage.  Just let the BIOS control it on AUTO.

P3D is a CPU hog and always has been.  Like I said, mine goes to 90% load at times while flying, but it does drop down after a minute or so.  I think it uses a lot of CPU power to load the scenery, and once it's loaded the usage goes back down for a while.

P3D sets my cgf affinity mask to use all 16 cores by default.  Even with the stock AF and all 16 cores used, I still get the high CPU load at times.  I changed the AF to exclude only core 0.

Dave

[Ray Proudfoot 9,058]

1 hour ago, dave2013 said:

I changed the AF to exclude only core 0.

That is extremely important. If you assign core 0 and core 1 to P3D you’re fighting with the OS.

[dave2013 1,158]

2 hours ago, Ray Proudfoot said:

That is extremely important. If you assign core 0 and core 1 to P3D you’re fighting with the OS.

Ah, so should I exclude core 1 also?

Thanks.

Dave

[newtie 541]

Dave-

This is the stock AM.

 

[JobScheduler]
AffinityMask=65535
P3DCoreAffinityMask=65535
MainThreadScheduler=0
RenderThreadScheduler=2
FrameWorkerThreadScheduler=4

Try this and see what the result is.

I have a 12900k and after hours of ******* around with various combinations and exclusions it’s working very well. Win11 does an excellent job of threading on its own. You can always go back.

[Bob Scott 7,774]

I just did the update from 5.3HF2 to 5.4 this last weekend.  CPU utilization does indeed seem to have taken a markedly different path in this version.

I have been running P3D with VSync on to a 30Hz display for years, which has limited CPU utilization on the heaviest tasked core running the main thread to down around 70%.  With 5.4, that core is pegged at 100% pretty much all the time, and the other cores (including the 10 e-cores enabled for P3D via the affinity mask) are also being fairly heavily tasked, especially when airborne.  So the temps and voltages are indeed averaging higher than before, but it's not causing a problem--in fact, the opposite.  It raised my eyebrows when I first saw it, but the sim is running very smoothly like this without any of the stuttering I would have expected from keeping the main core loaded up like that.  I just flew the FSL Scarebus into FlyTampa's KBOS, which has always been a challenge performance-wise, and it was very smooth despite the main thread's core being firewalled at 100% the whole way in.

I suspect that if your cooling isn't up to having most of your CPU's cores getting a real workout a lot of the time, it might cause CPU temp issues where 5.3 did not due to its much less aggressive usage of multithreading.

On my water-cooled 13900K, with 20 cores active and 18 allocated to P3D (AM=262143), I'm seeing temps venturing into the mid 70s C on the core running the main thread, which is about 8-10 deg C higher than I saw with 5.3.  Some of the other cores are occasionally bumping into the high 60s and low 70s.  The cores specifically allocated to the render thread and frameworker thread are not being tasked noticeably higher than the others like they used to...most of the cores are getting hit concurrently now, and fairly often.  It's anyone's guess as to whether the thread scheduler masks are now being ignored, or if their workload has been redistributed in such a way as to make their load levels indistinguishable from their peers.

 

[dave2013 1,158]

3 hours ago, newtie said:

Try this and see what the result is.

I went back to the stock AF setting and didn't really see any difference in performance.  I still get the periodic high CPU loads, but it doesn't cause any stuttering or performance drop when it happens.  Now core 0 runs at 90+% instead of core 1.

I read some posts in this forum which recommended setting the AF so that P3D does not use core 0, but it doesn't seem to make any difference.

Thanks.

Dave

[newtie 541]

That has been my experience.

[newtie 541]

@Bob Scott,

Are you locking externally?

Edited August 9, 2023 by Ray Proudfoot
Long quoted post removed.

[Ray Proudfoot 9,058]

8 hours ago, dave2013 said:

Ah, so should I exclude core 1 also?

Thanks.

Dave

I’m still on v5.3HF2 and given Bob’s experience with 5.4 I would be hesitant to make a recommendation.

[Juliett Alfa Romeo 567]

[JobScheduler]
AffinityMask=4095
P3DCoreAffinityMask=4095
MainThreadScheduler=0
RenderThreadScheduler=2
FrameWorkerThreadScheduler=4

Is this ok for i5 12400?

[SteveW 1,792]

2 hours ago, Juliet Alpha said:

[JobScheduler]
AffinityMask=4095
P3DCoreAffinityMask=4095
MainThreadScheduler=0
RenderThreadScheduler=2
FrameWorkerThreadScheduler=4

Is this ok for i5 12400?

That would be for an HT enabled 6 core CPU and this unfortunately allows scenery (and other objects) loading to share the MainThreadScheduler core 00 which brings excess heat and restricts the fps.

Instead I would use:

HT Enabled 6 core
05,04,03,02,01,00=core number
11,11,11,01,01,01=AffinityMask = 4053
11,11,11,01,01,01=P3DCoreAffinityMask = 4053
00,00,00,00,00,01=MainThreadScheduler = 0
00,00,00,00,01,00=RenderThreadScheduler = 1
00,00,00,01,00,00=FrameWorkerThreadScheduler = 2

This keeps scenery loading away from sharing the main thread cores 00, 01 and 02

So in Prepar3D.cfg:

[JobScheduler]
AffinityMask=4053
P3DCoreAffinityMask=4053
MainThreadScheduler=0
RenderThreadScheduler=1
FrameWorkerThreadScheduler=2

 

However, scenery loading can be increased by enabling the sister Logical Processors on cores 01 hosting RenderThreadScheduler and 02 hosting FrameWorkerThreadScheduler (which are not so active as the core 00 MainThreadScheduler) but these can be impacted from scenery loading depending on the system, so should be tested. This won't provide more fps but will reduce the time the CPU gathers objects and scenery as you fly. Giving:

HT Enabled 6 core
05,04,03,02,01,00=core number
11,11,11,11,11,01=AffinityMask = 4093
11,11,11,11,11,01=P3DCoreAffinityMask = 4093
00,00,00,00,00,01=MainThreadScheduler = 0
00,00,00,00,01,00=RenderThreadScheduler = 1
00,00,00,01,00,00=FrameWorkerThreadScheduler = 3

So in Prepar3D.cfg:

[JobScheduler]
AffinityMask=4093
P3DCoreAffinityMask=4093
MainThreadScheduler=0
RenderThreadScheduler=1
FrameWorkerThreadScheduler=3

 

The same theory to avoid core sharing with the main threads should be applied to any CPU using HT (or equivalent).

 

[SteveW 1,792]

An alternative to increasing scenery loading speed is to put both RenderThreadScheduler and FrameWorkerThreadScheduler on core 01. Again this depends on the system and weight of scenery and objects so should be tested:

HT Enabled 6 core
05,04,03,02,01,00=core number
11,11,11,11,11,01=AffinityMask = 4093
11,11,11,11,11,01=P3DCoreAffinityMask = 4093
00,00,00,00,00,01=MainThreadScheduler = 0
00,00,00,00,01,00=RenderThreadScheduler = 1
00,00,00,00,10,00=FrameWorkerThreadScheduler = 2

[JobScheduler]
AffinityMask=4093
P3DCoreAffinityMask=4093
MainThreadScheduler=0
RenderThreadScheduler=1
FrameWorkerThreadScheduler=2

 

[ttbq1 350]

1 minute ago, SteveW said:

An alternative to increasing scenery loading speed is to put both RenderThreadScheduler and FrameWorkerThreadScheduler on core 01. Again this depends on the system and weight of scenery and objects so should be tested:

 

HT Enabled 6 core
05,04,03,02,01,00=core number
11,11,11,11,11,01=AffinityMask = 4093
11,11,11,11,11,01=P3DCoreAffinityMask = 4093
00,00,00,00,00,01=MainThreadScheduler = 0
00,00,00,00,01,00=RenderThreadScheduler = 1
00,00,00,00,10,00=FrameWorkerThreadScheduler = 2

 

 

[JobScheduler]
AffinityMask=4093
P3DCoreAffinityMask=4093
MainThreadScheduler=0
RenderThreadScheduler=1
FrameWorkerThreadScheduler=2

 

 

Hi Steve, Any recommendation for 13900k? I am struggling with 5.4 also, any suggestion would be highly appreciated!

[SteveW 1,792]

With regard to the AMD 5800x3D with SMT enabled gives 16 Logical Processors from 8 cores we can use:

HT Enabled 8 core 16 LP
07,06,05,04,03,02,01,00=core number
11,11,11,11,11,01,01,01=AffinityMask = 65493
11,11,11,11,11,01,01,01=P3DCoreAffinityMask = 65493
00,00,00,00,00,00,00,01=MainThreadScheduler = 0
00,00,00,00,00,00,01,00=RenderThreadScheduler = 1
00,00,00,00,00,01,00,00=FrameWorkerThreadScheduler = 2

[JobScheduler]
AffinityMask=65493
P3DCoreAffinityMask=65493
MainThreadScheduler=0
RenderThreadScheduler=1
FrameWorkerThreadScheduler=2

 

With CPUs not using HT (or SMT) allow P3D to set the Affinity.

 

With regard to avoiding core 00 to allow for Windows processes this does not make a difference to P3D as Dave correctly suggests above.

 

[SteveW 1,792]

5 minutes ago, ttbq1 said:

Hi Steve, Any recommendation for 13900k? I am struggling with 5.4 also, any suggestion would be highly appreciated!

This is a 24 core CPU with 8 P cores and 16 E cores. The 8 P cores with HT enabled give 16 Logical Processors for a total of 32 LPs. I'll do the math for that and come back in a moment...