P3Dv5 is single core AGAIN 🤦

[OzWhitey 3,353]

3 minutes ago, vl1 said:

close this topic.

this is useless running water

You came here to rant - over and over - at LM, when it turns out you’ve got a problem with your AI. You don’t want to hear the obvious answer - turn your AI down! Be realistic!

Yes, this thread is pretty useless - it adds very little to the world. 

 

Edited April 23, 2020 by OzWhitey

[neumanix 113]

One thing I've learned is that neither CPU nor GPU should be pegged at 100% if you want a smooth and stutter free experience.

Another thing I noticed, is that setting the CPU affinity through the config file versus in the task manager gives different results.

The thing that has worked for me is to set the affinity in Task Manager TWICE.

First start the sim, then open Task Manager and the Details tab, find the prepar3d.exe, right click and choose Set Affinity.

Uncheck Core 0 and then OK. Now go back and do it again, this time enabling all cores again.

This seems to spread the load over the cores more evenly and should prevent one core from maxing out.

Don't know how and why this works for me, but try it and see if it helps.

Note also I don't have v5, still using v3 and FS9.

[cwburnett 1,809]

I noticed changed behavior from v4.5 to v5 with respect to core utilization and asked LM about it on the prepar3d.com forum.  Beau Hollis explained a bit about what the cores do in his reply:

Quote

To get an accurate test, you should be flying fast or slewing. The primary core is always working, as are two additional cores set aside for per-frame rendering related work. The rest of the cores are used for jobs like loading terrain, autogen, creating batches, etc. On higher setting, you'll see all your cores working hard whenever terrain is being paged in.

I would encourage any/all of you with experiences on this to contribute to that thread because it would be nice to have a better understanding of how best to configure our processing hardware to maximize performance.  https://www.prepar3d.com/forum/viewtopic.php?f=6312&t=137087

I had shared these two images with him:

Framerate unlimited, vsync to 30hz monitor:

Framerate limited to 30:

It continues to be clear that with framerates locked, we can expect the main thread (typically core0) to be fully saturated because the sim engine uses all the processor power available to both render the frames and perform tasks related to scenery loading (this is directly related to the many FFTF discussions over the years).  Meanwhile, with frames unlimited and using vsync or other external framerate limiter, the main thread only works as hard as it needs to based on the FFTF setting + the number of frames that need to be rendered.

I had raised the issue, however, because my cores 2 and 4 are never utilized above 50% (like @Sethos noticed as well), which is changed behavior from v4.5.  V5 is using net less processor power than v4.5 at the same settings and I was specifically wanting to be sure I wasn't missing something that could improve performance.  I'm guessing cores 1 and 3 are the ones Beau is referencing when he says "two additional cores set aside for per-frame rendering related work."

6 hours ago, GodAtum said:

Can anyone confirm if V5 likes cores or speed more?

I think "yes to all" is the answer here. It really needs at least 4 cores to do what it needs to do, but obviously the faster those cores can process, the better.  What I think doesn't help is having more than maybe 8-ish cores.  @SteveW has shared a lot on this subject and I've been a diligent student of his writings on affinity, here and on the codelegends.com forum as well.  In my case, I have 6 cores running at 5ghz and P3D isn't using them all to their potential, so I don't think adding more cores would buy me any more performance.

What does seem clear from my experience is that (a) @vl1 is correct that the main thread, typically found on core0, continues to be the primary driver of framerate from the CPU side of the equation and (b) that overall CPU usage is lower in v5 at the same settings as compared with v4.5, namely that in 4.5 I saw five of my six cores average above 80%, while now it is only 4 cores that exhibit that behavior.

[Boomer 1,619]

12 minutes ago, neumanix said:

One thing I've learned is that neither CPU nor GPU should be pegged at 100% if you want a smooth and stutter free experience.

Another thing I noticed, is that setting the CPU affinity through the config file versus in the task manager gives different results.

The thing that has worked for me is to set the affinity in Task Manager TWICE.

First start the sim, then open Task Manager and the Details tab, find the prepar3d.exe, right click and choose Set Affinity.

Uncheck Core 0 and then OK. Now go back and do it again, this time enabling all cores again.

This seems to spread the load over the cores more evenly and should prevent one core from maxing out.

Don't know how and why this works for me, but try it and see if it helps.

Note also I don't have v5, still using v3 and FS9.

Yeah I really dont want to have to do that everytime I fly, what a PITA.

[neumanix 113]

1 minute ago, Boomer said:

Yeah I really dont want to have to do that everytime I fly, what a PITA.

Well, it takes a couple of seconds so it's not that bad. Would be interesting to know if it's of any help in v5, so please try it and maybe report back.

[Doug47 574]

2 minutes ago, Boomer said:

Yeah I really dont want to have to do that everytime I fly, what a PITA.

Probably would've taken no more time then you spent replying to his message. 

[Boomer 1,619]

So I tried it, my 8600K is running all 4 processors at 100%?

[neumanix 113]

Just now, Boomer said:

So I tried it, my 8600K is running all 4 processors at 100%?

Ok, interesting. Did it change smoothness or anything? Maybe it's not working the same for v5.

[Boomer 1,619]

Just now, neumanix said:

Ok, interesting. Did it change smoothness or anything? Maybe it's not working the same for v5.

No change for me in v5, no.  Still a little stuttering.

[cwburnett 1,809]

3 minutes ago, Boomer said:

Yeah I really dont want to have to do that everytime I fly, what a PITA.

Agree.

16 minutes ago, neumanix said:

Another thing I noticed, is that setting the CPU affinity through the config file versus in the task manager gives different results.

The thing that has worked for me is to set the affinity in Task Manager TWICE.

I can't find it now, but this was asked and addressed on the LM site previously.  If I remember correctly, basically, this process breaks the way P3D manages terrain loading with locked framerate and core 0 then behaves, more or less, how it does with an external frame limiter - so core 0/main thread utilization falls to whatever it would be strictly based on FFTF and whatever processing capacity is required to generate the framerate, instead of using 100% of the capacity of core0.

This is exactly what I wish LM would explain better in terms of optimizing our machines. For example, in v4.5 I had success using process lasso to pull everything else off core 0, leaving P3D untouched. My experience was that IF I let process lasso touch the P3D process at all, my performance suffered, but by having it remove other processes from core 0, I had fewer stutters.  However, I've attempted this with v5 and have found that any use of process lasso has odd ramifications for P3D.  I'm still experimenting to see if I can work it out with v5.

[Sethos 4,319]

Nice to see you echo exactly what I've experienced @cwburnett

Especially playing around with AM / Process Lasso, which was fairly benign under previous versions, now has some profound effects. Doing anything to the mask through PL renders the sim completely unusable for me, everything takes seconds to load which freezes up the sim, several cores spiking etc. HT, something I never had any success with disabling through an AM, now also has a remarkable effect and makes the entire sim run much more efficiently and at higher FPS. Though, as you also note, I'm curious whether the lack of use on core 2 and 4 (can't remember if those were the exact same as mine or I had two different ones) is intentional or another bug.

In any case, lots have certainly changed in terms of how its utilizing the threads and I will definitely be using an AM that disables logical cores going forward.

[OzWhitey 3,353]

I’m keeping it simple, no affinity mask. Whether they help at all is debatable, and my impressiom from the LM team is that they haven’t designed the recent sims to run with one. So tweaking affinity masks may have helped a little for FSX, but i’d leave things be for v5.

Taking other programs off core 0 with process lasoo does sound reasonable.

[Bert Pieke 7,584]

For my flying thus far, I have been using an AM of 11111101 as I was for V4.5. and have seen improved smoothness..  like it a lot!   (quad core, HT on.)

Watching the individual core loads, I also see a different distribution with some cores lightly loaded.. I can only assume that this is intentional.  As a result, I have moved my GNS/GTN to the sixth core from the 3rd core, where I used to run it.

So far, so good.. 

[Sethos 4,319]

4 minutes ago, OzWhitey said:

I’m keeping it simple, no affinity mask. Whether they help at all is debatable, and my impressiom from the LM team is that they haven’t designed the recent sims to run with one. So tweaking affinity masks may have helped a little for FSX, but i’d leave things be for v5.

Taking other programs off core 0 with process lasoo does sound reasonable.

Can only speak for my own setup, but something is seriously messed up with HT under heavy load conditions, something I never saw under the previous iterations. HT on / off made no measurable difference, it was all based on feel and placebo for my own part. However, under v5 it's like night and day, with some very odd spiking and utilization of the cores with "HT on" i.e no mask, default conditions. Under "normal" conditions they both seem smooth but once you apply pressure with a full OrbX region (SoCal), flying out of Klax, a real machine killer, the cores just start to choke.

 

[cwburnett 1,809]

1 minute ago, OzWhitey said:

I’m keeping it simple, no affinity mask. Whether they help at all is debatable, and my impressiom from the LM team is that they haven’t designed the recent sims to run with one. So tweaking affinity masks may have helped a little for FSX, but i’d leave things be for v5.

Taking other programs off core 0 with process lasoo does sound reasonable.

There are really only three use cases for AM from my perspective - and I agree that when in doubt, don't use one.

1) If you have hyperthreading ON you don't want P3D's main thread to share a physical core with another P3D thread - that just cannibalizes main thread performance. Those with HT off, or without HT processors can ignore this.

2) If you use process lasso to manage processes. I found that NOT having an AM can cause problems where P3D doesn't put the main thread on core 0.  Again, this was just my experience and don't know if it applies to others.  I used the AM simply to force P3D to use all (6 of) my cores and it solved my moving main thread problem while using process lasso. Those not using process lasso can ignore this.

3) As @SteveW has advised, it is possible that at a certain point, more cores starts to degrade performance because of the process coordination involved, so if, for example, you have 16 cores, you could use an AM to steer P3D to only 8 of those cores, etc... Those with 8 or fewer physical cores can ignore this.

So, yea, most people can probably ignore AM.