CPU % - buffer size

I just installed the new 3.0.51 version on my old ‘music’ Macbook (2014 , Intel I7 2.2 GHz, 16 GB RAM, Big Sur OSX).

Tracking seems to work quite well, and I noticed a couple of things:
1) some synth use only 1 CPU, some other 2 CPU (as indicated in the upper right corner), and the ones with 1 CPU tend to cracle a lot, even thoug the CPU% seems not paricularly high.
When using 2 CPU I have not any problem.

2) My audio card is a Steinberg UR12 and I can use a 64 byte buffer size.
16 bytes is too low and crackles… does it depend on the Audiocard speed or my CPU performance?
I would like to try an intermediate buffer value of 32 bytes (not available now)…

I’d like to know other opinions, before thinking to get a newer and faster machine…

It’s tricky to explain. I have given this a lot of thought and try to make the best compromises, based on many years of experience. This is not an amp sim or audio FX, so forget everything you know about audio buffers.

  1. With “automatic threading” set in audio device settings, it will run on 1 or 2 cores, as it think it’s best. 1 core is less latency. 2 cores less CPU usage. You can enable “always multithreading” to avoid crackles with low buffersizes.

  2. Both, but mainly CPU. There is a diminishing return as you go lower. You probably wont feel a difference if you could run at 32, and the slightest CPU hiccup can make it crackle. But worse still, maybe the audio stream is fine, but the CPU core that is runs the tracking cant keep up and you wont know about it, and it degrade tracking quality.

Im not saying that 64 samples/buffer is good enough for everybody, and maybe 32 will be an option, but that there are a few things that needs to go hand in hand, and we need to optimise it as much as possible, and take the best compromises.

Btw, It’s also possible that the latency you seem to hunt is in another part of the system (audio interface or DAW).

To make sure I don’t run into any problems with a small buffer size, I play live at 128 samples and sometimes 256 samples and I’m not really bothered by latency.
Maybe that’s because I’m used to gigs where I sometimes play 6 or 8 metres away from my amp or wedges.
In any case, it gives me peace of mind… :wink:

This is not an amp sim or audio FX, so forget everything you know about audio buffers.

…so 64 or 128 byte buffer shouldn’t make such a big difference? I’ll try ASAP.
… and I’ll try also a Focusrite audio card, that probably has a bit better performances

one reason i’d like to see 32 samples is that there are times where i’d like to split the note detection from the sound generation.

these include: running large numbers of cpu heavy vst effects, playing percussion sounds and to improve midi bass note detection speed.

in this case i’d run mg3 standalone on interface1 at 32 samples and send the midi to my daw running on interface2 at 256 or even 512.

i did try to test this but although i can select 32 samples in the standalone config i think it’s still running at 128.

or maybe you’ve looked down this road and found it to be a dead end?

1 Like

I played a bit more, forcing the multithreading and things went better, more stable.
And actually also 128 was absolutely playable.

Also, there’s some work to be done for the player, to clean up the articulations and picking style.
I found myself using the usual muffled ghost notes in legato passages, or semipalm muted, that are so expressive with an analog sound, but completely dangerous for note detection.
A good motivator for working on technique. :smile: :smile:

2 Likes