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

[Bert Pieke 7,584]

4 minutes ago, Nemo said:

It appears that this 100% usage problem is known since quite some time.

It is not a "problem" but the way that P3D is coded..

[Nemo 1,305]

8 minutes ago, Bert Pieke said:

It is not a "problem" but the way that P3D is coded..

You are probably right, and I don't see it as a "problem" either as long as there are no performance issues and/or extreme overheating.

[Ray Proudfoot 8,930]

6 hours ago, gianluk81 said:

I tried the modification suggested by lm, but nothing at all ... this is my sad situation ... 100% fixed core and unnecessarily high temperatures. Among other things, I have been crashing to desktop continuously with the ActiveSkyUtils.exe module for two days, who knows if it is not related to this "gutted" use of core 0.

Do you have a large number of Ai injected into the sim? AIG for example.

[Nemo 1,305]

I've made a test  with some interesting results:

internal fps unlimited but with RTSS 1/2 vsync restriction (-> 30 fps): CPU 0 =95-100% other below 50%; GPU 80-90%

internal fps 30 (Vsync, TB ON) (-> 30 fps): CPU 0 =100% other below 50%; GPU 80-90%

internal fps unlimited no other restrictions (-> 40 fps) : CPU 0 =50-65% other 40-50%; GPU 95-100%

Since this was only one test in a rural but demanding scenery, I cannot conclude much. But imo it indicates that internal fps setting and further restrictions on or off play an important role. What I'm wondering the most, that unlimited fps with no further restrictions gave the most balanced CPU usage. I had expected it the other way around. But as @Ray Proudfootsuggested, AI injections may be crucial as well, as in my test not much ai was present.

Edited June 15, 2021 by Nemo

[JoeFackel 760]

Guys ... its not the case that processor power just can be shifted between CPU and GPU. Thats not how this works.

If there is higher CPU load than it may have to do with the VRAM optimization all had cried for. Now the CPU must shove more data from and to the VRAM. There is no graphic processing that has shifted by magic to CPU on 5.2.

 

[Ray Proudfoot 8,930]

10 minutes ago, Nemo said:

What I'm wondering the most, that unlimited fps with no further restrictions gave the most balanced CPU usage. I had expected it the other way around. But as @Ray Proudfootsuggested, AI injections may be crucial as well, as in my test not much ai was present.

Limiting fps within P3D has always carried a penalty. Hence why VSync came about and using your monitor refresh rate to control things.

From the suggestion by LM the change will only affect how Ai operate so without a significant number you probably won't see anything different. Hence why I conducted my test at EGLC. Lots of nearby buildings and two big airports within 30nm. Probably the most severe test available.

7 minutes ago, JoeFackel said:

If there is higher CPU load than it may have to do with the VRAM optimization all had cried for. Now the CPU must shove more data from and to the VRAM. There is no graphic processing that has shifted by magic to CPU on 5.2.

Sounds like they have moved some processing back to the CPU from the GPU. But it's all gone onto Core0 which was already high whilst the remaining cores have a much reduced workload. A better spread would be ideal but I guess that isn't possible or they would have already done it.

But the underuse of the GPU is frustrating. My 1080Ti is only using 1/3 of its capability.

[JoeFackel 760]

2 minutes ago, Ray Proudfoot said:

Sounds like they have moved some processing back to the CPU from the GPU. But it's all gone onto Core0 which was already high whilst the remaining cores have a much reduced workload. A better spread would be ideal but I guess that isn't possible or they would have already done it.

Like i wrote, you can't just move processing between CPU and GPU. These are two VERY different animals.

[gianluk81 3]

2 hours ago, Ray Proudfoot said:

Do you have a large number of Ai injected into the sim? AIG for example.

I do not use the traffic ai, I only have the mtl of ivao and flai (which however I have always used on p3d), but I have already done some tests excluding both mtl ivao and flai, but nothing, this disproportionate use of core 0 remains.

[Ray Proudfoot 8,930]

4 minutes ago, gianluk81 said:

I do not use the traffic ai, I only have the mtl of ivao and flai (which however I have always used on p3d), but I have already done some tests excluding both mtl ivao and flai, but nothing, this disproportionate use of core 0 remains.

Understood, thanks.

I conducted two tests. One with the parameter set to True and another with it set to False.

Located at EGLC with two major airports to the west and south the number of Ai on the Ground for True was 79.

The False setting gave 173 on the ground. But for both CPU0 remained at 100% more or less continuously on a flight from EGLC to EGLL with clear skies.

It appears to make little difference apart from the number of Ai that remain but those have to be deleted via FSUIPC to maintain 30fps.

[craigeaglefire 197]

1 hour ago, Nemo said:

internal fps unlimited no other restrictions (-> 40 fps) : CPU 0 =50-65% other 40-50%

like what I am seeing at heavy airport like FSDT KORD with plenty of TG AI aircraft - also FPS unlimited…

with Active Sky live weather,

@4K resolution

[gianluk81 3]

I also noticed that, at least in my case, it doesn't even affect the fact that you use unlimited fps without any external limiters. Always 100% core 0. Has anyone tried, by chance, to see the use of cores with active hyper threading? So out of curiosity. I don't have it active in my 9900kf.

[IanHarrison 521]

For what it's worth, my 6700k @4.4 has been manipulated with Process lasso to push most  of the other programs off core 0/1 (HT on).

I am getting nearly 100% on all cores and GPU usage of about 40%.

This at EGLL with AI and TE GB south. RTSS doing the limiting to 30fps,which is almost steady.

[craigeaglefire 197]

5 minutes ago, IanHarrison said:

with Process lasso

also using Process Lasso with 9700K which only has 8 cores…

[micstatic 4,316]

I did a quick flight last night (actually a day flight) SFO-LAS on VATSIM.  It was actually a fairly quiet night traffic wise.  No other AI.  I noticed some microstutters here and there.  Especially at Vegas.  This never happened before I updated to 5.2.  For the update I completely uninstalled 5.1.  Then completely reinstalled 5.2.  Also made sure both shader folders were gone and that I started with a new p3d.cfg file.  I continue what I've done since installing V5.  Monitor set to 30hz.  Then in the sim unlimited frames / vsync on / triple buffering off.  I'm going to try alternate ways of limiting my frames.  Or potentially try not limiting them at all.  But I can say as of right now I'm disappointed with this update.  Performance has degraded and in my opinion with very limited enhancements.  Especially for somebody who doesn't even like EA that much in the first place.  High chance I roll back.  

[neilhewitt 493]

This is a long one, but I thought I could take a moment to put the '100% on a core' thing in perspective. 

100% on a core all the time could be a strategy for ensuring that the core is only used by your code. Like I said up-thread, you can consume all the CPU time on a given core without doing anything useful. If I write the statement:

while (true) ;

This will loop forever doing absolutely nothing, and it will consume 100% of the core it runs on. If I then change it like this:

while (true)
{
    if (some_condition)
    {
        // do some useful work here
    }
}

Then this will still loop forever, but whenever some_condition is true, it will do some work and then go back to 'sleep'. It'll still occupy 100% of the core.

If I write this code, however:

while (true)
{
    if (some_condition)
    {
        // do some useful work here
    }

    Thread.Sleep(1);
}

That last call to Thread.Sleep() will tell the Windows scheduler 'I'm done for now, you can schedule other work on this core' and another thread (maybe from my program, maybe from another) will get an opportunity to run. The core may still be at 100% if the next thread has work to do, but if all threads have nothing, or at least nothing intensive, then the core uses drops below 100%, potentially all the way to 0% (which doesn't actually indicate zero work, there's always a base level of work going on by the scheduler but that is removed from the number you see).  

I have no idea how LM has actually written the code for the main thread, and since not everyone here reports 100% use of their cores, clearly it's not as simple as the above situation. But the point I want to make is that 100% usage of a core does not necessarily indicate something 'bad'.

Also, unless you have entirely inadequate cooling, running one core at 100% even 100% of the time can't damage your CPU. Running all cores at 100% should not be able to damage it either. To heat-damage the CPU, the thermal overstress would need to occur quickly enough to prevent thermal throttling (which pulls down the core speeds to bring the temp within a safe range) kicking in; generally, that could only happen if you've massively over-volted the chip (naive overclockers do this sometimes) or have no cooling at all. If you overclock a chip that has a huge overhead available (like, say, an i7 8th-gen boosted to 5GHz from a base of 3.2GHz) then the thermal load increases significantly, and if you run all cores at that speed then your CPU temp will go very high very quickly, but still, thermal throttling will kick in to protect it. It's very very hard to damage a CPU just by running it hot and fast; you'd have to change BIOS settings that are generally locked, so it would have to be intentional.

As JoeFackel mentioned above, GPU vs CPU is not a simple question of 'moving work'. They are very different types of processing core. Some work that used to be done on the CPU was moved to the GPU in a different form (tesselation) and since then a bunch of extra work has been added to enable new graphical capabilities in the sim at little CPU expense. It may well be that some work around allocating VRAM for DX12 has been added to the CPU in this release to ensure that VRAM is allocated and released quicker, thus reducing the overall VRAM load, but this would not massively increase the amount of work the CPU has to do. 

I don't honestly see this as a problem. I appreciate some of you will feel differently, but that's my opinion based on 25 years as a commercial software developer and a simmer for longer than that. 

(That's not to say I don't think it's a problem if 5.2 performs worse on the same hardware as 5.1... but with some people reporting performance decreases and some significant performance increases, it's clearly not as simple as that.)

Edited June 15, 2021 by neilhewitt