Kontakt 7 is causing high CPU usage on Mac Studio M1 Max

@drkpttrn Consider creating a separate thread for your issue in the following section below:

I'm following this with interest as I'm also on a Mac Studio (M1 Ultra) and some similar hardware to @Intersolar (go RME!) but haven't yet had unusual crackle issues.

If others can provide more examples of patches in libraries (preferably NI ones others are likely to own like Pharlight) and even MIDI files, I'll try it on my system too to see if I get overloads. Especially if it's with a single patch.

I am seeing that if I try Pharlight - The Alchemist while playing dozens of voices at once or doing a glissando, then I do get system overload. But that seems excessive so I'm hoping to repro it with something that's maybe just a few voices, even a chord.

Also in my case, I'm currently NOT using Kontakt multitimbrally, since that distributes multicore load better, from what I understand.

I too see Ableton CPU usage shoot up to 30%-ish when playing a few chords. But if I duplicate the instance multiple times (and manage to not crash), it looks like it holds at 40%-50% with 16 instances and isn't otherwise being unruly.

I'd expect all M1s to perform similarly in single-threaded.

I would expect these patches to not be an issue on an M1, especially if they worked fine on an older Intel.

Ableton said the same thing about buffer sizes. I tested Pharlight after reading this thread, and there were a couple of things I noticed:

• Kontakt 6 and 7 behave the same with the same patch. I don't notice any difference in performance. I haven't tested against my old Intel Mac but it was so underpowered compared to the M2 Pro I don't know that any valid comparison can be made. I ran that on max buffer for everything and it sometimes still had channel overloads with instruments based on Reaktor Blocks and Pharlight.
• Going down below 512 increases CPU load shown in Live, and how Live behaves. By the time I get down to 128 or 64, that channel is overloaded and the instrument is glitching. I had the same behaviour with Vocal Colors. Ableton even say 'In some case with certain CPU intensive processes, lowering the buffer size will not reduce CPU load. In that case it might be better to higher the buffer size.' So the advice boils down to an unhelpful 'lowering it might help, increasing it might help, we don't know.'

My experience has been that if I am working with anything CPU intensive in Live, I need to be on minimum 256 buffer. Some instruments like Pharlight and Vocal Colors only function well at 512 and above. One Vocal Colors preset had me going to 1024 before it stopped glitching.

Where things get weird is I checked CPU use in Activity Monitor (which has a window where you can see all cores) and Live's per channel CPU for the channel which overloads on low buffers. CPU use, efficiency vs performance, is not much different between 128 and 1024 samples. Both show large amounts of unused capacity. And about the same use of efficiency vs performance (slightly higher efficiency on one setting, not enough to mean much). Yet at 128 the channel is overloading while no core is overloaded. And at 1024 the channel still shows 50%+ usage, yet no core is even close to 50%.

This leads me to believe that audio software is operating inconsistently, and Ableton in particular has some serious issues to work out on M1/M2 processors – in this case it is maxing out a channel before a single processor core is even at 50%. I'd love to understand how performance vs. efficiency core decisions are made when running a process. I assume, on no evidence but seems a safe bet, this is not under control of the audio software, but is selected by MacOS based on what it thinks should be used.

It would be nice if a process could say 'use performance cores for everything until they're almost at capacity, only then offloading some lower CPU operations to the efficiency cores' or the switch from efficiency to performance could be more predictable. Perhaps this will happen over time.

I've yet to see evidence that using buffer sizes over 256 pushes processing onto the efficiency cores – Live consistently shows lower per channel use on higher buffer sizes and MacOs monitor shows very little difference – slightly more efficiency core usage, but performance cores still in use. I'm not saying it's not true under some circumstances. I'm saying the evidence of my own measurements has not supported it and I've not seen evidence posted online, only the claim that it happens. My current practice is: when using anything which is very high CPU like Pharlight or Vocal Colors, buffer at 1024. Otherwise 256 or lower. This has worked so far.

----- Images showing Live and Mac performance measurements -----

In these images, the leftmost 4 cores are efficiency, the other 8 are performance. The Live channel is playing a one note at a time melody, no chords, on a Vocal Colors preset.

Mac and Ableton monitoring at 128 buffer:

Mac and Ableton monitoring at 1024 buffer:

I am disappointed with how poorly Live performs on the M2 Pro in certain situations. On the other hand, I had a Live Set using the Toybox Tangle Synth which on my previous core i5 4-core Intel Mac could only play one instance, single note melody not chords, at a time. Loading a second channel with another Tangle Synth glitched the whole set unless I froze one of them. That same set plays, all 4 Tangle Synth instances unfrozen, without problem, at lower buffer size, on the M2 Pro. So I kept duplicating the Tangle Synth until I had something like 12 - 16 of them and it still played without glitches! MacOS showed a lot more CPU use than the above images: all cores were up to around 75-80%. Yet Live kept playing fine and showed relatively low load per channel (total load was high, as would be expected). So whatever's up with certain demanding instruments and this inconsistency in how Live performs with them (channel maxed out, glitching) and what MacOS shows is happening (very low use), overall this machine is lightyears ahead in audio performance.

It is possible the problem is in Kontakt, or these specific Kontakt instruments. Or it's in Live. Difficult to say. I've had one other plugin obviously misbehave, overloading on the M2 Pro when it used almost no CPU on my Intel Mac, but that was in one set and I can't replicate it in any other.

I haven't found anything which allows disabling efficiency cores on MacOS. It may be possible, but if so the mechanism is currently either unknown or not widely known. What would be ideal for me is a way to set certain applications to never use them, but let the OS use them, since most OS tasks – at least those still running while making music – won't need performance cores.

This StackExchange post has something to say about whether and to what extent the software developer can affect which cores are used: https://apple.stackexchange.com/a/426930.

The M2 Pro has only one chip, with 12 cores, so we can rule out possible intercommunication problems between chips.

Adding to my last post, I forgot to look at Kontakt's CPU meter. It is doing even wilder things than the Live channel meter. It hops between around 15% and 25% much of the time, with semi-regular swings to 100%. It also goes red a lot, presumably indicating an overload, but it does this at both 100% and 25%. The behaviour is much too erratic to capture in a screenshot. But it is interesting that the meter provided by Kontakt provides a third diverging view of CPU use. So we have three views that each disagree on the exact load at a given time, except in that two of them are saying there are overloads, with the OS saying there is no problem, more than enough CPU headroom while the software says different and glitches.

I realise that Activity Monitor does not provide continuous real time CPU, it has an update frequency. I had it set to 2s but retested now with 1s and saw no differences. Is it possible the spikes occur between updates? Sure. Consistently? When watching for an extended time? Unlikely. The spikes are not evenly spaced such that they could always be hidden from the Activity Monitor view. And Live's meter is consistently high – never dropping more than 1 bar from full – while Kontakt's goes red at a high rate.

It's a mystery, and one that needs people who know the internals of the software in use, much more than I do about how these chips work, and have access to the code of the applications so that they can analyse what's happening – i.e., NI and Ableton – to figure out.

I just updated to Kontakt 7 so I could add Fables to my collection (the demo was amazing) but it's useless with the CPU load, it overloads even when Kontakt is the only plugin loaded and I'm playing just a single note! I have a 2019 Intel based Mac Pro and have never had any issues on it to date for any Kontakt instrument or any other plugin for that matter, until loading Kontakt 7.

If you are on 7.6.1 try downgrading to 7.6.0 that fixed this issue for me. See this thread for the download link for 7.6.0: https://community.native-instruments.com/discussion/17691/kontakt-7-6-1-update-causes-cpu-spikes-crashes/p1