1 Jun
2010
1 Jun
'10
9:53 a.m.
On Tue, Jun 01, 2010 at 12:18:57AM +0200, Robin Gareus wrote:
On 05/31/2010 10:42 PM, fons(a)kokkinizita.net wrote:
This is a bogus argument.
Keeping software buffers does not help to reduce the number of
times the CPU has to wake up - this depends *only* on how much
the audio *hardware* is able to buffer. For this reason chips
designed for portable use have something like 2 seconds of
internal buffering.
320 MB corresponds to more than ** 13 minutes ** of stereo signal
in floating point format at 48 kHz.
If an audio subsystem is using 320MB of memory
it's broken. Just dump it.
I think you followed Lennart's presentation @LAC2010 [1] explaining the
motivation for this: Pulseaudio keeps buffers sometimes 10 secs long or
even more. The reason for that is to have the CPU wake up as little as
possible in order to save power. FWIW jackd also locks ~80MB in /dev/shm here.
The computer I'm writing this on has just 512 MB. ATM it's
playing back 32 channels in Ardour, and processing them
with jconvolver. I still have 300 MB free.
Ciao,
--
FA
O tu, che porte, correndo si ?
E guerra e morte !