Buffer Size

[ali_b]

Hi, When I set the buffer size to 160/192/224 the plugin gives me a yellow triangle warning “Run your host at 128 or 256 samples at 44.1/48kHz - or 256 or 512 samples at 88.2/96/192kHz. Other buffer sizes cause extra latency.”
(Curiously when I set buffer to 64 it does not.)

Do the buffer sizes between 128 and 256 cause MG2 problems (at 44.1/48kHz)?
(and are bigger buffer sizes any better for pitch tracking?)

There does not appear to be an adverse effect on CPU measured by the host or on MG2’s meter at any of the buffers between 128 and 256.

thanks
Alistair

[in case it’s relevant:-
MBP15 mid 2012 MacOS 10.13.6 , Presonus FireStudio Mobile or Presonus Quantum 2626 (new toy!) running Gig Performer as host. 224 seems to be the sweet spot for not hearing/feeling latency on my Quantum - being less than 11ms RTL. At 256 (12.1ms) I get a nagging sense of ‘something’. (Monitor system is adding nearly 2ms to those numbers) I like the idea of more cpu headroom for the plugins better than ultra low latency I can’t percieve.]

[fogle622]

Computers work more efficiently when data is transferred as a multiple of a binary number like 2, 4, 8, 16, 32, 64, 128 or 256. If a data message falls between those numbers it gets broken down into pieces that add up to one of those numbers.

So when you set a buffer size at 160, 192 or 224 you’re not allowing the computer to work as efficiently as it can.

[ali_b]

Thanks fogle622
They may not be powers of 2 but
160= 5x32
192= 6x32
224= 7x32
so they are multiples of a sensible binary friendly size data chunk.
The devs will know if that is acceptable or if multiples of 64 would be better. Or indeed just the powers of 2.

Al

[fogle622]

Al, I agree computers can handle them. There’s no disagreement between us on that point. You stated in your first post you stated there was no visible difference in cpu use so the odd buffer sizes do not visibly affect performance.

My statement was the odd buffer size likely affects efficiency. I doubt anyone can tell if the difference between 99.5 percent and 100 percent efficiency. Small changes in efficiency may not affect performance but can have a cumulative effect. For example Paul indicates he tests with six instances of MG2 running. Do odd buffer sizes make a difference then? Perhaps but I don’t know.

[Paul]

MG uses FFT (Fast Fourier Transform). This transform converts the incoming audio into a spectrum. The FFT algorithm needs his input buffer sized as power of 2. That means: 2,4,8,16,32,64,128,256 etc
If you choose an in between value, MG has to shuffle the data a bit to retain the power of 2 requirement.
There is no harm in using those in between sizes though: we just warn for the extra CPU it might cause on weaker machines.

[ali_b]

many thanks for the insight Paul

Al

[fogle622]

Al,

That was a great question. I’m glad you asked. Now we know the warning is by design but not necessarily an indication of an issue.

Thanks for asking!