[PD] latency, -frags, and -blocksize

jfm3 jfm3 at ouroboros-complex.org
Mon Jan 14 16:28:53 CET 2002


Hmm. OK. Then how can I measure sound card latency? My present 
understanding is that I run with the lowest number of frags at the 
smallest blocksize possible before I start hearing dropouts in the 
audio. I run with -noadc and don't get any difference in the lowest 
numbers I can get at. There must be more smarts to it than that. Can 
someone clue me as to a better process for getting these variables set 
right?

(jfm3)


Miller Puckette wrote:

> Hi, one small comment...  "blocksize" only controls the I/O block size,
> not the DSP block size, which is always 64.  Control latency is always
> 64 samples but timing accuracy of I/O for control depends on I/O block
> size, since timing is all derived from ADC and/or DAC transfers which are
> made each I/O block.


jfm3 wrote:
>>Thanks for your reply.
>>
>>I set -noadc (since I don't really use the audio in anyway, just 
>>prerecorded material). The lowest I could get it was
>>-frags 8 -blocksize 64
>>[...] Then, at that point, removing -noadc made no 
>>difference.


Andrew (Andy) W. Schmeder wrote:
>>>blocksize is the dsp vector size, also controlls the temporal resolution
>>>of messages... ie. your midi messages can only be processed on block
>>>boundaries.  Midi having a clock with 1 ms resolution means you want a
>>>small blocksize to match.  bs = 64 => ~1.4 ms resolution which is
>>>decent.
>>>
>>>frags is latency... i.e. the number of blocks of lag.  IMHO its better
>>>to have more frags than a large blocksize.
>>>
>>>I have an i8x0 sound card in a laptop as well (tosh portege 3440ct) and
>>>with lowlat patch and alsa 0.9b.  Basically the card sucks.  In full
>>>duplex its impossible to to run with less than 15 ms latency... i.e.
>>>blocksize 64, 10 frags.  However if I disable the dac or the adc (-nodac
>>>or -noadc) then I can get it down to 1.5-3 ms latency.... but usually
>>>that's not desirable.





More information about the Pd-list mailing list