Has MacOS Sonoma finally fixed real time threading issues on Apple Silicon?

[Tone2 Synthesizers]

Thanks for the info. What macOS versions exactly were affected by it?

[Tone2 Synthesizers]

Never had a single problem either, until recently, and only on Apple Silicon (Intel have been working like a charm). I think it happened when they introduced the Audio Workgroups mentioned earlier. From what I have read on the Internet, it broke for many others. I guess it indeed depends on how thread pools have been implemented, but there is no (good) reason why it would break suddenly with a new OS release...

Anyway the problem seems to be gone with the latest OS. Knock on wood.

Thanks for the info. We'll keep an eye on it. What macOS versions are affected by it?

[Blue Cat Audio]

Thanks for the info. What macOS versions exactly were affected by it?

Not 100% sure I remember well, but it was at least with MacOS 13 (codename Ventura). I am 100% sure it was fine on Big Sur, and I think I recall having no trouble on Monterey too, at least with the first iterations.

[Richard_Synapse]

I upgraded yesterday and see no difference in MT performance compared to any previous OS X. Did Apple claim improved MT performance in its press release somewhere?

Richard

[Blue Cat Audio]

It technically not an improvement of MT performance but rather some changes in the scheduler that make MT for real time audio work again. I don't know if this is caused by a change in how threads are assigned to performance cores or simply if priorities are properly managed (when a real time thread waits for other threads, whatever their priority). I don't think this is the kind of stuff that Apple would advertise anywhere, just like when they finally fix graphics performance after several releases of regression.

[Tone2 Synthesizers]

I assume that things got messed up a bit with the introduction of the efficiency cores. Similar things happened on PC with the new AMD processors as long as Windows did not natively support it. It was fixed with an OS update.

[Blue Cat Audio]

I assume that things got messed up a bit with the introduction of the efficiency cores. Similar things happened on PC with the new AMD processors as long as Windows did not natively support it. It was fixed with an OS update.

It is indeed likely connected to the efficiency cores. However in this case Apple manages both the software and hardware so they could make sure that they are kept in sync... And also this issue was not present with earlier versions of the OS.

So I guess they introduced some changes in Ventura to improve energetic efficiency, which caused all sorts of problems, which have (maybe) finally been fixed with the release of Sonoma. Let's cross fingers that this won't break again in the near future!

[aMUSEd]

Corel Painter 2023 has a built in performance benchmark tool, when I bought my M1 Studio Max I was so excited to get 'awesome' performance using the tool but when I upgraded to Ventura I was extremely disappointed to find it dropped all the way to 'good' - nearly 2,000 points lower.

I nagged Corel for months to update it to fix the performance issues with Ventura, they eventually released a patch but it only had marginal effect on performance. However I have just installed Sonoma and ran the tool again and am now seeing a really huge improvement:

Screenshot 2022-12-05 at 20.16.33.png

[Blue Cat Audio]

Corel Painter 2023 has a built in performance benchmark tool, when I bought my M1 Studio Max I was so excited to get 'awesome' performance using the tool but when I upgraded to Ventura I was extremely disappointed to find it dropped all the way to 'good' - nearly 2,000 points lower.

I nagged Corel for months to update it to fix the performance issues with Ventura, they eventually released a patch but it only had marginal effect on performance. However I have just installed Sonoma and ran the tool again and am now seeing a really huge improvement:

That's interesting! However if you can trust the report details, it looks like the performance improvement is mainly related to the GPU?

[Blue Cat Audio]

Well, it looks like I was actually wrong and there is no magic with the scheduler in MacOS Sonoma!



The observed "new" behavior was actually a side effect of a bug in the AppKit framework (read the article for more details). It confirms that when Apple Silicon CPUs are busy, they perform better!

[mystran]

The observed "new" behavior was actually a side effect of a bug in the AppKit framework (read the article for more details). It confirms that when Apple Silicon CPUs are busy, they perform better!

It's probably a CPU throttling thing: when the system sees that the average load isn't too high, it probably drops the clockrate 'cos ... hey that's enough on average and then when you have a real-time thread that doesn't care about averages, but rather absolute time to complete a task, it's screwed.

[Blue Cat Audio]

The observed "new" behavior was actually a side effect of a bug in the AppKit framework (read the article for more details). It confirms that when Apple Silicon CPUs are busy, they perform better!

It's probably a CPU throttling thing: when the system sees that the average load isn't too high, it probably drops the clockrate 'cos ... hey that's enough on average and then when you have a real-time thread that doesn't care about averages, but rather absolute time to complete a task, it's screwed.

That's indeed what I thought. Too bad! Maybe I should just have a process that keeps all CPUs as busy as possible all the time .

[mystran]

The observed "new" behavior was actually a side effect of a bug in the AppKit framework (read the article for more details). It confirms that when Apple Silicon CPUs are busy, they perform better!

It's probably a CPU throttling thing: when the system sees that the average load isn't too high, it probably drops the clockrate 'cos ... hey that's enough on average and then when you have a real-time thread that doesn't care about averages, but rather absolute time to complete a task, it's screwed.

That's indeed what I thought. Too bad! Maybe I should just have a process that keeps all CPUs as busy as possible all the time .

Then you risk the opposite problem where the system might decide cores are too hot and throttle to cooldown.

[machinesworking]

The observed "new" behavior was actually a side effect of a bug in the AppKit framework (read the article for more details). It confirms that when Apple Silicon CPUs are busy, they perform better!

It's probably a CPU throttling thing: when the system sees that the average load isn't too high, it probably drops the clockrate 'cos ... hey that's enough on average and then when you have a real-time thread that doesn't care about averages, but rather absolute time to complete a task, it's screwed.

That's indeed what I thought. Too bad! Maybe I should just have a process that keeps all CPUs as busy as possible all the time .

Years ago there was a bug in a particular Intel chip that Apple had in their laptops that throttled down when not asked to perform. It hit the audio world with things like a complex sample library in my case Akoustic Piano causing a glitch in Live the first time a song was played through when that part came up, but not subsequent times. At the time someone wrote a little script that poked at the cores keeping them active, without actually using CPU, that fixed it until Apple added something similar in an update.

[Blue Cat Audio]

Years ago there was a bug in a particular Intel chip that Apple had in their laptops that throttled down when not asked to perform. It hit the audio world with things like a complex sample library in my case Akoustic Piano causing a glitch in Live the first time a song was played through when that part came up, but not subsequent times. At the time someone wrote a little script that poked at the cores keeping them active, without actually using CPU, that fixed it until Apple added something similar in an update.