Hello, for example, when I play vigorously on Keyscape and use the sustain pedal extensively, sometimes I hear a crackling / audio dropouts after a while. No matter how I set the buffer size, with the latest hardware, 64 GB of RAM, NVMe, etc. It can be reproduced on both notebook (Intel I7-155H) and desktop (Intel I7-265K, Win11 64 GB). The CPU runs smoothly at 20%, and no red CPU spikes are detectable. When I turn off multiprocessor in the audio menu, the problem occurs less frequently. It was already the case in the older versions. In the task manager, during the first half of the time, it is visible with multiprocessor and in the second half without. Essentially, the core usage is very uneven, and the developers should work on this because a stable system is essential for me as a cover band musician. But just my question: Why is the performance WITH multiprocessor worse than without? Thank you in advance.
Since there is already a topic about multiprocessing, I will write here.
I have such a problem in this version: multiprocessing spontaneously turns off in the settings. I have already seen this four times in 2 days.
I think I have found a solution, I have switched off the e-cores on both machines (and left multipr. off). Currently there are no more glitches. Seems to be more due to the e-cores. Sometimes fewer cores are more…
It is quite a complex problem.
One thing to note is that no matter how good or bad multiprocessing works, your plugin that has to cope with the sustain pedal can not be distributed across cores. That is one audio process, and those cannot be distributed, they are monolitihic.
Furthermore, only processes that don’t depend on others can be processed seperately. Say you have a channel with 2 sends to 2 other channels, then the latter cannot be processed before the sending channel has done its work. Thus, they cannot be processed in parallel on multiple cores. So the system has to figure out which processes are independant, which most of the time works great if you have myriads of Layers or Stacks which can run independently, have no sends in the mixer etc.
This all comes with a caveat, which is that there is some overhead to organize all those processes which needs to be done for each buffer processing. They need to be protected from interfering with others which also means there are some wait functions idling. So unless your configuration is such that it benefits from MP (multiple singular, monolithic processes), the outcome may even be worse than processing without.
It’s too complex to provide a one-stop receipe, you just have to try and possibly analyse your routing if it can be optimized for that, but it is complex.
That is why I tested deeply with audio and video multithreading, and in the end the absolute best performance I have (lots of stacks, lots of sends, lots of outputs) is with audio AND VIDEO both multiprocessing OFF !!
probably because of that…
Maybe make MP a song by song option?