On Mon, 2008-09-01 at 23:13 +0200, Paul Davis wrote: > On Mon, 2008-09-01 at 22:39 +0200, Paul Davis wrote: > > On Mon, 2008-09-01 at 23:29 +0200, Jens M Andreasen wrote: > > > > IMO, 10 - 30us latency is fantastic, but perhaps a little over the top > > > for audio work. > > > > at 96 to 192kHz SR, not really. 16 samples at 192kHz is about 8usecs. > > or even 83.3usecs :) >
That's better ... > but 20usecs latency would still imply stealing 25% of the period ... > Right, but only if you use a period of 16 samples. Don't know about you, but I run into severe performance problems with so tight loops and datasets exceeding cache level one. Not to mention that I currently try to offload to the gpu, where the data disappears into a black hole, not to be heard of again before some 250us later. The thing is that if you convert your measure of latency to distance rather than to an abstract percentage, even a full millisecond does not sound so bad. After all, it's only some 330 millimeters, or about the same distance as from your ear to the guitar you are (or might be) holding. I used to sit on my amplifier, which means the speaker must have been at least a meter away, giving a latency of about 3ms. Don't know how I could manage? :) _______________________________________________ Linux-audio-dev mailing list [email protected] http://lists.linuxaudio.org/mailman/listinfo/linux-audio-dev
