Firstly - what a great piece of software ZynAddSubFX is!

Only problem I have had is, under linux, is if I patch 4 channels from Rosegarden to 4 ZynAddSubFX channels in the same process/instance, the output glitches a *lot* whilst playing - audibly and in Audacity if I record the mixed output.

Glitches are only a few mS but repeat every couple of seconds or so.

The system is not busy other than the audio processes and there is plenty of idle time on the CPUs.

Mixing is done via Jack and jack-mixer.

The sound hardware is driven with ALSA, the system has 6GB RAM and dual core Intel i3 M370@2.4GHz


Now - the funny thing is - if I run 4 instances of ZynAddSubFX and route each Rosegarden track to a different ZynAddSubFX, then mix via jack-mixer, it plast flawlessley, every time.

This does not feel right - and I suspect there is a good reason why asingle instance is stuttering on multiple channels.

Any ideas?

Cheers

Tim

Which version are you using, and do you have a RT kernel?

Hi,

2.4.0-1.2 - and I have a "low latency" Mint 13 kernel 3.2.0-36

As far as I'm aware, that's just compiled with HZ=1000.

I presume you mean the PREEMPT_RT patches? If so, no, not yet.

I just tried the whole setup on another laptop booted from USB stick to "Ubuntu Studio 12.10" - again, no RT kernel (well, not that I am aware). Same problem.

I *wonder* if the way the threading works mean that 4 separate instances can migrate to the 4 virtual cores (pair of hyperthreaded cores) whereas in single instance mode the threads cannot make effective use of the cores???

Cheers,

Tim

Hmmm. They are currently up to 2.4.3 so that might be the problem, I remember that the early 2.4.0 versions did have glitch problems.

These days I use the Yoshimi derivative and only occasionally check out Zyn. itself. However, if you wait around I'm sure one of the devs will drop in

{try shouting a bit}

> {try shouting a bit}

It seems to work fine.

> 2.4.0-1.2
I'd recommend trying out the more recent version, as there were some fixes made that could make things run more smoothly.
I also tend to avoid any distro subversions with my software as that implies that they were fooling around with the source, so I have no clue what '-1.2' implies without figuring out the patches applied to the source.

As you noted having multiple instances should help even out the load.
IIRC Jack 2 (and not jack 1) distributes processes to different cores when the signal graph permits, so that should help if you are using expensive patches.
It is quite an undertaking to get a single process to distribute its own load safely so neither zynaddsubfx nor the derivative mentioned by folderol do so.

My bet is on you overloading one core with the single instance, though I may be wrong.
Getting RT stuff to work reliably on a system is quite an interesting affair (from kernels, to interrupts, to permissions, to message queues, to bit fiddling, to optimizations, etc).

> I just tried the whole setup on another laptop
Well, if you are interested in staying with the debian/ubuntu family, there is
http://kxstudio.sourceforge.net/

I have not tried it personally, but it seems to have a decent reputation from what I have seen.

Thanks for the tipoff - Yoshimi takes 4 channels without a problem.

It's a bit prone to crashing if you cycle through the instruments too fast, but it is glitch free.

BTW - just found one of your sound banks "Collection" - nice set

Best regards,

Tim

Hi,

I can do that - I'll build a patch free from lastest deb package and see how that goes.

Many thanks,

Tim

2.4.3 seems a lot better - I'll hit it hard this weekend and see how it goes.

Thanks for the tips folks