P3Dv5.x core usage notes

[Mace 1,927]

So I wanted to post this before P3Dv5HF1 dropped, but I decided to wait until the hotfix was released.  Some of you have already noticed this, too -- I am seeing different core usage than I did with P3Dv4.x, and this is with a completely stock/vanilla/unmodded sim (i.e. no AffinityMask).  A picture tells 1,000 tales:

Notice the interesting thing.  Core2 and Core4 seem to be a mirroring Core0.  In other words when Core0 has a spike in usage, so too, are Cores2 and 4.  Clearly, the other cores are showing the tell-tale signs of being terrain loading cores, and we know what Core0 is, but what is the story with Core2 and Core4?    And is this particular distribution unique to the 6-physical-core 8700K cpu?   What are you all seeing with your core usage (you 8 physical core guys, in particular)?

[Bert Pieke 7,585]

I have masked off Core1 on an HT quad core CPU, and instead of Core2 and 4, my low usage cores are 3 and 5.

There is definite pattern to Core loading, similar to yours, just shifted one position to the right.

[cwburnett 1,809]

Lost in the depths of the prepar3d.com  forum from a few weeks ago is my question about this and Beau's (from LM) response...

https://www.prepar3d.com/forum/viewtopic.php?f=6312&t=137087&p=213761

You can see from my graph I've got the same behavior. It appears that with frames unlimited at 30hz I basically can't saturate my core 0 anymore and by experimenting with my add ons, I've found that now core1 is the most sensitive, so I am now experimenting with steering all add-ons and processes outside P3D to cores 2 and 4 and also to 0,2,4.

 

 

[DavidP 1,412]

Sounds like too much trouble to me. I just let it do its thing.

[Mace 1,927]

2 hours ago, cwburnett said:

Lost in the depths of the prepar3d.com  forum from a few weeks ago is my question about this and Beau's (from LM) response...

https://www.prepar3d.com/forum/viewtopic.php?f=6312&t=137087&p=213761

You can see from my graph I've got the same behavior. It appears that with frames unlimited at 30hz I basically can't saturate my core 0 anymore and by experimenting with my add ons, I've found that now core1 is the most sensitive, so I am now experimenting with steering all add-ons and processes outside P3D to cores 2 and 4 and also to 0,2,4.

 

 

Ah thanks for the thread there.  So Beau says there are (now, with v5) two additional cores being used for "per frame rendering work".  That explains what I'm seeing.

Right now I use a hexadecimal of FFC for all add-ons affinity (assigned via batch file), which keeps add-ons off Core0.  I think instead, based on your findings with Core1, I'll now limit them to some later cores...like 10 and 11 or 8 thru 11.

To your post in the LM thread, I haven't yet experimented with totally saturating the main thread.  As you can see in my pic above though, I'm only hitting about 70% usage.  That will climb as I do my Heathrow testing, night testing, complex aircraft testing, etc.  I would agree with you, that if you push the main thread to 100% yes you run the risk of a stutter.  At least in v4 that's what we saw.

What interests me are these two extra cores Beau Hollis is talking about.  And how they seem to mirror the main thread.  This may account in part for the smoothness in v5.  Maybe rogue processes that would spike main thread usage, are now intercepted by these other two free cores, and the work is done there, avoiding an overwork on the main thread. (?)

4 hours ago, Bert Pieke said:

I have masked off Core1 on an HT quad core CPU, and instead of Core2 and 4, my low usage cores are 3 and 5.

There is definite pattern to Core loading, similar to yours, just shifted one position to the right.

Thanks for the data on a 4-physical core cpu.

[lownslo 592]

6 hours ago, Mace said:

...this is with a completely stock/vanilla/unmodded sim (i.e. no AffinityMask).

 

Just to clarify... you are running no flightsim addons beside the sim during your testing?  Also, could be usefull if we knew what non-sim related apps are running in Windows while P3Dv5 is running.

Thanks,

Greg

[Mace 1,927]

1 hour ago, lownslo said:

Just to clarify... you are running no flightsim addons beside the sim during your testing?  Also, could be usefull if we knew what non-sim related apps are running in Windows while P3Dv5 is running.

Thanks,

Greg

No addons, nothing else running.  That was the point, to test without anything else, and observe the core usage.  I'm curious if you see the same thing with your 6-core.

[DonMuskopf 32]

8 hours ago, Mace said:

Core2 and Core4 seem to be a mirroring Core0. 

I find the same pattern on my 8 core CPU with hyper threading enabled, except that my Core0 (i.e. LP0) is at 100% utilization. LP2 and LP4 are definitely doing something different from the other LP's. LP2 and LP4 seem to be helper threads for LP0.

I'd be curious to seem what the pattern is with HT off. The first three cores working together?

[cwburnett 1,809]

1 hour ago, DonMuskopf said:

I find the same pattern on my 8 core CPU with hyper threading enabled, except that my Core0 (i.e. LP0) is at 100% utilization. LP2 and LP4 are definitely doing something different from the other LP's. LP2 and LP4 seem to be helper threads for LP0.

I'd be curious to seem what the pattern is with HT off. The first three cores working together?

I don't have HT on, six physical cores. Same pattern. 0,1,3,5 used. 

Core0 runs at 100% when framerate locked or unlimited without vsync. I'm at 30hz unlimited vsync on and core0 is 60%-80%

What's new for me in v5 is that core 1 always runs near 100% and cores 2 and 4 never go above 50% no matter what I throw at it.

 

[DonMuskopf 32]

25 minutes ago, cwburnett said:

core 1 always runs near 100% and cores 2 and 4 never go above 50%

All of this very interesting. So it seems that if you have 6 or more cores, the pattern for the main P3D process is to use Cores 0, 2 and 4 (or with HT on, to use LP0, LP2 and LP4).

[Mace 1,927]

7 hours ago, DavidP said:

Sounds like too much trouble to me. I just let it do its thing.

That's the point David, I am just letting it do its thing...no mods at all.   What is too much trouble?

By itself default / vanilla P3Dv5HF1 is reserving 2 additional cores for extra work.  All I can say is, the results seem to be good.

[lownslo 592]

4 hours ago, Mace said:

No addons, nothing else running.  That was the point, to test without anything else, and observe the core usage.  I'm curious if you see the same thing with your 6-core.

I've not moved to v5 yet.  Your results and others posting about the additional rendering cores/LPs bodes well for HT.  I started using v4 years ago with a HT quad core, and was able to make good use of HT.  Once I moved to my current 6 core the HT capability was less useful, so running HT off was the obvious choice for me.  But now LM has brought better balance to how cores/LP's are utilized by the sim, making HT more important to the "more cores" crowd.  Thanks LM! 🥂

Greg

 

[Bert Pieke 7,585]

15 hours ago, DonMuskopf said:

All of this very interesting. So it seems that if you have 6 or more cores, the pattern for the main P3D process is to use Cores 0, 2 and 4 (or with HT on, to use LP0, LP2 and LP4).

Simply observing the Performance tab in Task Manager shows that these cores are not being used for texture/scenery loading, so one can assume they are being used by P3D for its main processes...  Whatever the magic sauce is, V5 is smoother on my system than V4.5. which I used to say was very smooth already

[ShezA 176]

V4.5: I have been running some experiments on P3DV45 and Process Lasso. My problem is heat on GPU/CPU. I was on HT off for a time because of this.

I recently went back to HT ON and I found that my win10 performance improved significantly.

I then went back to PL and set up a profile which when turned on moved P3D to HT OFF (it is in the CPU affinity drop down). I moved all other processes to the other cores.

Result: Lower temps and same performance and better Win10 performance.

Very satisfied!