Subject: Re: [linux-audio-dev] a quick note about latency expectations
From: Tom (stillone_AT_snowcrest.net)
Date: Thu Dec 06 2001 - 23:43:07 EET
> >16 frames per cycle! yow! this h/w can really do that?
>
>
> you can even go down to 2 frames/cycle but obviously everything below 64
> is not really practical with my single-cpu 450 MHz system. 16 can run
> in perfect sync for a few seconds until the kernel in one of its ignorant
> moods decides to block the audio thread for a while :(
Since the number of frames/cycle directly determines the amount of time
the computer has for performing rt dsp, there is probably no reason to
want to go below 64 frames.
However, if you are simply interested in recording without performing
dsp, the lower the better. I have just begun using a fostex D1624 hard
disk recorder. It has 2 adat ins, 2 adat outs, 8 analog ins, and 16
analog outs (similar to the original design of ardour). By
pre-recording to track 1 then routing analog out 1 to analog in 2, and
adat out 1 to adat in 2, then playing back track 1 while recording track
2 from the analog input and track 9 from the digital input, then
transferring the whole thing into my daw, I measured 87 samples of
latency on track 2 and *20* samples of latency on track 9. This is a
true full duplex experiment. Either fostex has implemented capture
offset or they use 10 frames/cycle. The fact that latency on the
digital ports was >0 suggests that they don't perform capture offset.
Tom
This archive was generated by hypermail 2b28 : Thu Dec 06 2001 - 23:38:44 EET