Zebra 3 Teething Troubles

[Funky40]

My point to be made:
In general was my observation that a M4 CPU is crackling much faster than a M2 CPU compared to the shown CPU load in %. And more plugins seem to be prone to force the M4 CPU into crackling (this one is vs. FX in my case)

.........because in no way should an M4 perform worse than an M2, and that makes me think even more your issue is the DAW and OS not communicating well and your track getting stuck on efficiency cores.

ohh, you overlooked here something.
please see what i`ve set in bold here.
it was "vs. %".

i`ve edited out all the rest / not required as a direct respond.

[tumface]

Something is wrong with your setup. Probably your DAW. Sorry, it's a bit hard for me to understand your posts. I have an old M1 MacBook Pro, and I haven't found any factory presets that overload a CPU core. (I just went through about 100 of them playing 6 notes at the same time, and I didn't have any overloads.)

[Funky40]

Something is wrong with your setup. (I just went through about 100 of them playing 6 notes at the same time, and I didn't have any overloads.)

No, it`s not.
You too, you overlooked information that has been written right there:

Also:
i do run my M4 mac with Zebra3 -on one core- into full cPU overload cause a patch ( this is the equivalent of one track in a DAW) consists of ALOT more than just Zebra3 alone.

Nevertheless, i have some Great presets in Zebra 3 that i can play freely without any CPU problems. While some presets ( original presets) run the core "immediately" to its knees.... (EDIT: same here: it`s in context to my whole patch. This is not Zebra3 standalone....)

please read that again / i`ve just highlighted a 2nd part


i quoted that part now here, while i delete the bigger part of that post afterwards, just leaving the necessary info vs. what`s necessary for the previous answer.

[Urs]

If M4 has worse performance than M1 then it most likely tries to run the plug-in on an efficiency core of the M4. This could be the case if the host doesn't schedule the plug-in rendering thread with proper Quality of Service attributes whatsoever. It could also be an issue with the audio interface / driver.

[Funky40]

...........then it most likely tries to run the plug-in on an efficiency core of the M4.

i personally never had any situation that gave me the feel that my host (gigperformer) was running on an E-core. True with M1, M2, M4 macs (desktops). To notes: GP is running "by itself" single threaded only. (which means: it`s the ever same CPU load situation)

- - - - - - - - - - -

Please read this: (this might be important info to be aware of)
There has been a guy or a group of guys -unaffiliated with apple- who wrote a testroutine vs. how the M4 CPU works. Their results where quite interesting. The M4 CPU works completly different than a M2.

A M4 CPU is permanently switching cores ! Even just running a single threaded app ( or especially then / idk)
The time intervals we`re talking here are "very small". This is not seconds. This are -small- fractions of seconds. (really very small fractions)


edit: i`ve not found a link

- - - - - - - - - - -

Even with patches just loading GP at 35% or so would it always run on a P- core.
I´ve really never seen GP falling apart cause it would run on an E-core.
I can`t speak for DAWs,...nor the other guy/guys with high CPU load problems with their M4 macs.

in my case, the host, the usecase, the workflow, everything, is day by day same !
I work often enough on patches i started years ago on a M1 or M2 mac.
In my case, I`ve never seen CPU anomalies P-core vs. E-core.


This whole "E-" vs. "P-" core situation "could" look btw. different on M5 macs ! with their CPU architecture

If M4 has worse performance than M1 then it most likely tries to run the plug-in on an efficiency core of the M4.

the two guys who responded to me, did both overlook parts of what i said.
It created in their brain a picture that looked like my M4mac was running worse than their M1mac.


Several completly diffent factors came all of a sudden together to one.
When i run my CPU into overload, do i know why this has been.
( The point of my post has originally been: Some -Great- sounds work perfectly, while some others have the CPU explode / as has been the case with what others have said)



YET, ...and this is a point in this case...... "audio crackling" on a M4 CPU* ( * as reported also by others) does not necessary mean it was just CPU overload. This very topic can be much more complicated vs. M4 mac CPUs.

sorry for long posts ! just trying to be informative

[eclipse_soundlab]

Removing mod assignments in modmatrix (x button) changes the mod amounts for other assignments in the patch on random.

It can also change (temporary - it will reset on next (any) mod assignment) mod connections.
example - when i drag little mod-cob on filter 1 (Picture 1) - comb1 ratio mod amount is moving in the mod matrix (Picture 2)

* Which operating system?
* Which host (DAW)?
* Which plugin format (VST3, CLAP, AU, AAX)?
* Can you share a preset example?

Windows 10 22H2
Reaper 7.1.60
VST3
Try factory presets with a lot of MSEG via mod matrix.

If version of reaper matters i will update (im not home, using laptop, RME UFX interface and old evaluation version of reaper as VST host)

[eclipse_soundlab]

This video shows deleting bug that triggers another bugs:

https://streamable.com/5o5smt

and this wrong conntection bug:

https://streamable.com/w7lvll

As you can see in second one EQ parameter is controlled by MSEG 2 (in the mod matrix) not Filter 1 (it should be filter 1 and MSEG 1)

used factory preset Janglebell Pad and added few MSEGs to mod matrix.

[uOpt]

So, let's say you have a VST instrument that's sitting idle and barely doing anything, because it's not having any notes played into it. The thread it's on, owned by the DAW, might be sent to an efficiency core by the OS. Then, you throw both hands onto your MIDI keyboard and play 8 notes at once, the CPU usage from the VST instruments spikes because it's now synthesizing 8 voices at once, up from 0 voices, and the OS decides to move that thread onto a performance core.

Moving the thread takes time and stalls for a moment, so you get an audio dropout.

I don't think so. Moving a thread between cores, even different kinds of cores, should be fast enough not to cause an audible glitch. That is easy to verify on non-Mac platforms by using taskset to force such a move.

You might get an audible glitch because of an efficient core overload before the move happens, that is a much more likely explanation for an artifact you might be observing.

[tumface]

Not sure what you mean. Migrating threads is probably few dozen microseconds at least, accounting for caches and stuff. If you don't have a lot of runway on the efficiency core, it's not going to be able to move the thread to a performance core and then finish the rest of the work in time. This is easy to verify in a DAW that has this problem on Apple Silicon, like FL Studio prior to 2026.1 beta 3, where you can observe this happening without any of the cores reaching full throughput. (They added workgroups API usage in beta 3, fixing the problem.)

[ffx]

The icon of this button is not clear. My first thought was this is a download button for an update, like in Discord. But it is a Save button, normally this symbol should be a floppy disk.

I would prefer something with lazers, maybe a prisma lazered disc. Floppy makes Zebra 3 instantly looking outdated. In the end, I am fine with the current symbol.

[tlvid]

When loading the attached preset which I created and playing E above middle C (E3) I'm getting a horrible oscillating ringing noise in the background. I've done the same set-up in Serum and Pigments and it's not occurring. Is this normal? https://www.transfernow.net/dl/202604233QYGLbp6

You get this only on this one note (E3)?
The ringing which I see in the output is coming from the band-limiting, I guess.

Screenshot 2026-04-24 212107.png

Sorry for the late reply. I'm not entirely sure what's happening, as I can hear the unpleasant high pitched sound on E notes but no others (on the factory init preset so basic saw wave only), and now I'm hearing the same in Serum and Pigments too, whereas I didn't hear it last week. It must just be my amateur ears (although I have no hearing issues).

[uOpt]

Not sure what you mean. Migrating threads is probably few dozen microseconds at least, accounting for caches and stuff.

The CPU caches' contents are not moved over, they are rebuilt from scratch.

There's no way that the move takes multiple hundreds of a second. As I said, it can be tested on audio on non-Mac platforms. Just force the DAW and all children on a set of efficiency cores, then use taskset to force a move to a set of performance cores and see whether that crackles.

[tumface]

The CPU caches' contents are not moved over, they are rebuilt from scratch.

Yes, and that takes time.

Just force the DAW and all children on a set of efficiency cores, then use taskset to force a move to a set of performance cores and see whether that crackles.

That's not equivalent, because you are causing it to happen at a (effectively) random moment instead of when the pressure on the CPU core has suddenly increased.

Instead of an artificial test on a different hardware platform and different kernel, you can see what actually happens on macOS on Apple Silicon when a DAW does not use the workgroups API by using FL Studio 2025. It will screw up and underrun quite frequently if CPU usage of an instrument fluctuates up and down, even if the CPU load never comes close to fully loading a core. This was recognized as a problem and fixed in the newest beta versions.

Sorry, I don't want to stay off-topic in this thread any longer. That's all I have to say about this.

[tasmaniandevil]

This video shows deleting bug that triggers another bugs:

https://streamable.com/5o5smt

and this wrong conntection bug:

https://streamable.com/w7lvll

Thanks, I can partly reproduce this now.
Deleting a mod slot changes the slot numbering. But the mod depth slider is still connected to the previous slot number.
This seems to be a visual UI bug, i.e. it doesn’t change the sound of the preset.

[tumface]

When deleting a mod slot, the 'via Depth' values of the mod slots that move to the left will take on whatever value of the mod slot that was there previously.

Screen Shot 2026-04-30 at 02.38.42.png

Screen Shot 2026-04-30 at 02.38.46.png