CPU spike- Catalina / 2020.5

[dmitch]

Originally Posted by Darryl Ramm

(And importantly it does not only matter if you use any plugin in your session or not... we care if you have removed all third party plugins installed on your system (and potentially all Avid plugins if you have incorrectly upgraded Pro Tools)... plugins don't need to be instantiated in your session to cause all types of problems, they just need to be installed in Pro Tools).

That is surprising to me. What kinds of problems can an installed plugin cause if it's not instantiated? Does plugin code actually run when it's not instantiated? That would be appalling. Can't PT get the info it needs from a plugin from some static config file?

Seriously, I'd like to understand more about this. Thanks.

[Darryl Ramm]

Originally Posted by dmitch

That is surprising to me. What kinds of problems can an installed plugin cause if it's not instantiated? Does plugin code actually run when it's not instantiated? That would be appalling. Can't PT get the info it needs from a plugin from some static config file?

Seriously, I'd like to understand more about this. Thanks.

That should not be a surprise, it is always how it has been. There must be thousands of threads on DUC talking this, including examples of people finding this problem, either with CPU type errors or memory leaks with uninstantiated plugins.

A plugin is not just an inert piece of code. It gets to run and perform initialization while being loaded into Pro Tools. You hope that code does what it should. There is absolutely no guarantee it does.

[Darryl Ramm]

Originally Posted by dmitch

What? Why? There is no difference at the application level between 2 cores running hyperthreaded and 4 cores running without hyperthreading. The meters could make sense if they were programmed correctly.

No, there is a huge difference, you need to read up about hyperthreading and think about what is going on.

hyper-threaded cores don't even really exist. They rely on internal unused capacity on the real core, if that real core is fully busy its hyperthreaded twin would never get to run. And both cores cannot be 100%. So what then are the meters even showing when showing hyperthreading? ... and "good performance" looks like what on those meters?

"meters programed correctly"... what technically do they actually show? Presumably its based on clock driven sampling. What is the sample rate? How do they avoid scheduling synchronization producing skewed results? What were the changes that supposedly improved the meter accuracy a few releases ago. Where is any of this explained? I doubt anybody here knows any answers to this, and I doubt most folks in Avid do either.

Just stop trying to overthink this, disabling hyperthreading especially can help, but start by just doing more standard troubleshooting and try to see if things get stable. You don't seem to have tried to fully exclude plugins, and there is no description of wether you have disabled dynamic plugin processing or not... if that can help is dependent on the plugins, and while it has become less useful to disable as Pro Tools has improved I sure would still test the effects.

And right now all this is not really helping the OP make progress.

[dmitch]

Originally Posted by Darryl Ramm

No, there is a huge difference, you need to read up about hyperthreading and think about what is going on.

Actually I wrote multithreaded code for many years when I was in the Advanced Computation Group at Apple. It was one of my fortes. I am thoroughly familiar with this area of software engineering.

hyper-threaded cores don't even really exist.

They are called virtual cores. Does a virtual core exist? Does virtual memory exist? Does a virtual machine exist?

They rely on internal unused capacity on the real core, if that real core is fully busy its hyperthreaded twin would never get to run.

That is definitely not true. If both virtual cores have work to do, they will both run.

And both cores cannot be 100%.

They absolutely can. One virtual core running at 100% means that it's running non-idle-loop code 100% of the time it's scheduled to run. Both virtual cores associated with one physical core can absolutely be running at 100% of their allotted time.

So what then are the meters even showing when showing hyperthreading? ...

The meters in PT? Today? They are showing nothing useful. But check out the meters in Activity Viewer. 4 physical cores, hyperthreading on, they show 8 cores. Well-written concurrent code can peg all 8 meters at 100%. I've seen this many times. Never with Pro Tools, though.

and "good performance" looks like what on those meters?

See above. Max performance is the meter for each virtual core showing a lot of activity, approaching 100%. I never see this with Pro Tools. I see it with Photoshop all the time when looking at the CPU meters in Activity Viewer..

what technically do they actually show? Presumably its based on clock driven sampling. What is the sample rate? How do they avoid scheduling synchronization producing skewed results?

The kernel's scheduler knows which virtual core is running at which times.

What were the changes that supposedly improved the meter accuracy a few releases ago. Where is any of this explained? I doubt anybody here knows any answers to this, and I doubt most folks in Avid do either.

Well I agree with that boldfaced remark, and I think that is one sorry state of affairs. If the folks at Avid don't know how these meters work or what they show, why are they there?

[Darryl Ramm]

Originally Posted by dmitch

That is definitely not true. If both virtual cores have work to do, they will both run.

They absolutely can. One virtual core running at 100% means that it's running non-idle-loop code 100% of the time it's scheduled to run. Both virtual cores associated with one physical core can absolutely be running at 100% of their allotted time.

You are confusing yourself with the illusion of a core presented by software. The actual execution units of these additional virtual cores do not exist, two hyper-threaded cores cannot produce twice the same work as two physical cores. Yet the meters imply they can/do. That implication is just wrong. A four physical core system and a 4 virtual core system is comparing apples and a duck. None of this has any relevance to virtual memory or virtual machines.

(and from the other post, "the scheduler knows what is running" is not helpful, of course it does, at any instant, and the CPU meters are sample based and don't know what is running outside of those samples. It is not hard to get into sampling/strobing effects, maybe more so when trying to do pseudo real time stuff. and I am not aware of specific issues there, but I also just do not trust the meters. IIRC from past discussion on metering one of the long since departed Avid engineers was thinking of writing up explanations on metering, unfortunately he left before doing that).

But none of your advice here is likely to help the OP. If you are having problems with your system maybe start a new thread and describe what troubleshooting you have done and somebody might be able to help you.

And to be clear... I am not defending Avid here or claiming their plugin runtime does not have issues. I am trying to avoid people going down time wasting rat holes looking at stuff that is unlikely to help them actually solve problems. If there is say a performance regression on 2020.5 or 2020.5 only on Catalina then lets find it, ideally without third party plugins, lets get other users to confirm that and lets hold Avid accountable for fixing it.

[RGrierson]

Try using the Renice command in the terminal app
this helped with playback for me! I didn't notice any difference in the CPU meters but Protools performed much better on my 2010 Mac Pro

I use a value of -16
so you would type sudo renice -16 then the PID number
Learn how to find the PID number below:

https://morganhvidt.com/how-to-prior...ons-on-mac-os/

[BCnSTL]

Just one person’s experience here: I have a 2018 MBP - 6-core i9 - I disabled Hyper-Threading and Turbo-boost and now my system runs very stable. I haven’t seen a -9173 error since doing it. Might be worth trying.

[Southsidemusic]

Originally Posted by BCnSTL

Just one person’s experience here: I have a 2018 MBP - 6-core i9 - I disabled Hyper-Threading and Turbo-boost and now my system runs very stable. I haven’t seen a -9173 error since doing it. Might be worth trying.

Well worth a try as ”turbo” boost and HT doesn’t make PT run any faster or better

[Davo Deluxe]

Hey I was having this issue rn and tried everything, until literally all I did was disable the “ignore errors” thing under the hardware buffer. Instantly fixed it and can run at a 32 buffer. It was constantly spiking at 100% every second, not a single issue now. I hope this is the fix for anybody looking for one


Sent from my iPhone using Tapatalk

[rockafella]

Originally Posted by dmitch

The meters in PT? Today? They are showing nothing useful. But check out the meters in Activity Viewer. 4 physical cores, hyperthreading on, they show 8 cores. Well-written concurrent code can peg all 8 meters at 100%. I've seen this many times. Never with Pro Tools, though.

See above. Max performance is the meter for each virtual core showing a lot of activity, approaching 100%. I never see this with Pro Tools. I see it with Photoshop all the time when looking at the CPU meters in Activity Viewer..

DMitch - I read this thread as part of my research into this issue. I have an 18-core i9 build with 128 GB of RAM, quad-SSD Raid 0, 1080TI, and on HDX. Yet with just a few HDX plugins, I noticed that usage was quite a bit, and then, just with one instance of Synthmaster, that raised it even more, and putting that Instrument track into record, brought it to 100% quite often and made it stop playback frequently. (not even in recording, just playback with record enabled on it)

I respect Ramm's thoughts and he's had plenty of great messages on here, but I agree with you, Pro Tool's CPU meter is garbage. Probably legacy code from a long-gone era of single-core CPU's, for all I know.

I was willing to give it the benefit of a doubt in that at least maybe it's just representing the one core with the most usage, but obviously one instance of Synthmaster doesn't max out a core (and all the rest of the stuff is on the HDX as DSP plugins).

So, I tried a lot of the usual fixes. Wasted the whole day trying different combos of my 300+ plugins to see if any seemed like the culprit - the only real trend I saw is, the more plugins in the Avid Plugins folder, the higher the "baseline" load, even on a near-empty project such as mine. I did seem to find some benefit from setting priority to background processes, not really sure; also kept the change where I disabled Indexing on the whole Avid Program Files folder. Now at least, even though it shows red/100% often while having the Instrument track in record, the loop will play back for a few or more minutes without stopping, so hopefully one of those changes did help a little bit. (can even keep my 30+ browser tabs open, etc.)

I might try an external USB video card just to see if that helps. (since, as we all know, the Nvidia driver is "hungry" and holds its interrupts on the PCIe bus longer than it really ought to)

Now, I finally tried the last thing - I can't BELIEVE no one has tried this (it fits with the weird established "fixes" of disabling cores, disabling turbo, disabling hyper-threading, etc.) --- PLEASE TRY THIS! Simply, run any CPU intensive benchmark. For example, I ran Cinebench R20 (All cores, mind you - not the single-core test).

As SOON as it started to REALLY max out my CPU - GUESS WHAT - the Pro Tools CPU (with my test project playing back still, and the Instrument track record-enabled) went from 90-100% to -- get READY for this -- 6 PERCENT! Yes, 6% - and with all the 36 individual cores' meters going down from their previous 10-100% range to all being 2%, with MOST being 1%!

So, I think it's safe to say - unless your WINDOWS Task Manager shows your CPU is getting hit hard in a big session, or, you know, you actually HEAR pops/crackles/dropouts - it's safe to just click that Ignore box in the Sound Setup and forget any of this ever existed.

I do like the fact that Pro Tools TRIES to pick up on any chance of a click though... but they really need to update their CPU monitoring to make it any good. I would be willing to bet that it might simply be related to the number of samples/length of time they measure over. I.e., if they took more measurements, and only threw the error up if a second or two passed with what they think is high usage... but, just as, or more likely, as you said - they simply are not counting the cores and dividing it out properly. In fact, perhaps they are somehow ADDING the usage from all cores, making the number even larger than what it would be (rather than simply taking an average, median, or max core CPU usage value).

Try my test though of loading up the CPU for real while playing back in Pro Tools. Again, I cannot believe no one has done this! (at least not from any searches I did, and I think I read almost all threads from the past five years on this, even ones before that timeframe :) ) I have also now tried running the single CPU test, and while it doesn't push it down anywhere near as much, it does make it much less likely to throw the error and stop playback.

So the real solutions seem to be:

1) Click the ignore box, and shake our fists at Avid for wasting so much collective time on a meaningless meter;

2) If you really want the meter to be more accurate and keep the behavior of Pro Tools stopping if it thinks you're at high usage - try to find a multi-core CPU stress tool that lets you set how hard it hits it. (would be kind of ridiculous to pummel it at a true 100% load on all cores just to see the meter register lower!) I think AIDA64 and others do this, maybe I'll try it sometime. I'll probably just click that checkbox though, because I really don't see a way that HDX with i9 18-core will ever have an issue for the software and plugins that exist today. :)