[ardour-users] Latency Trouble

[Jesse Chappell]

Christopher K. George wrote on Wed, 07-Apr-2004:

 > I am running Ardour 0.11.2 on latest testing version of Debian using a low-
 > latency patched kernel from Agnula.  I compiled the latest versions of ALSA 
 > and JACK myself.  I am running an Athlon 2600 with 500MB of RAM.  I have a 
 > separate HD for recording audio that supports sustained data rates of 60MB/S 
 > according to hdparm -Tt.  I am using an RME HDSP Multiface.  I have assigned 
 > it IRQ 9 which is not shared with any other device.  I have 
 > set /proc/sys/kernel/lowlatency to 1.  I have disabled all services that I do 
 > not need.  My week point may be running the latest testing version of X and 
 > KDE.
 > I recently had to change mother boards.  The old one had some issues plus I 
 > couldn't assign PCI IRQs.  On the old system I had to use a frame size of 
 > 8192 in order to avoid xruns.  With this setting I was able to record eight 
 > tracks indefinetly.  On the new MB I am able to function with a frame size of 
 > 512.  From the testing I have done so far, the system runs fine with that 
 > setting.  I am starting jack as root with:
 > 
 > jackd -R -d alsa -r 96000 -p 512.

Thats about 10ms roundtrip latency, which isn't too bad, really.

 > I can record all eight tracks with a frame size of 128 for a while, but then 
 > an xrun will happen.
 
How about '-p 256' ?

 > Any suggestions?  I am getting along alright using hardware monitoring but 
 > would love to have some better latency.
 
Unless you really need to monitor your input with host-based effects
or you play live sw synths or samplers, why do you feel the need
for lower latency?  Not only does it increase the possibility
of xrun's that ruin the precious recordings, it also increases
CPU overhead.  The HDSP hardware matrix mixer is excellent for
zero-latency monitoring (you are 'making do' with it) assuming you don't need
the items mentioned above.

If you do, in fact, have the real need to get extremely low-latency
response, it sounds like you've done all the right things, and
you may have hit the limits for your hardware and kernel.  Chances
are, any further improvements to be had lie in kernel version/patches
and scheduling.

jlc