On Wed, Mar 15, 2023 at 06:51:20PM +0100, Robin Gareus wrote: > Very nice plots. Well almost, even though it's in the description I'd expect > axes to have labels :) :-) > In case of P = 96, N = 3, do the 5 outliers cause x-runs No. > Have you calculated the average clock speed for all settings? It is the same in all cases. I tested this when I developed an app to log mains frequency some years ago. The sample rate error doesn't matter much for the measured frequency, but it does for the timestamps if you want to record this for e.g. 24 hours at exactly on second intervals. My plot for P = 48 and Giso's for P = 32 suggest there is something happening at below 1 ms resolution. The question is where and how this is done... The driver could use a timer and some extra buffering to 'fake' a shorter period. But there would be no latency advantage doing that... Ciao, -- FA _______________________________________________ Linux-audio-user mailing list -- linux-audio-user@xxxxxxxxxxxxxxxxxxxx To unsubscribe send an email to linux-audio-user-leave@xxxxxxxxxxxxxxxxxxxx
