Hi, > Just a shot in the dark here, but Skype had an issue with PulseAudio 4 That has been fixed in the last release. > According to the screenshot, the time is spent in the ppoll syscall. > This rings a bell: your bug may be a duplicate of an already-fixed thing. Thanks for the pointer. Before I go through the trouble compiling upstream PA, the bug-report mentions PA starts to poll in a spinning loop consuming 100% cpu - while I see PA consuming 25-30% with 10% CPU load still available. > First, the algorithms are more complex. Second, X server doesn't do much. > 640x480 pixels * 30 fps * 2 bytes per pixel = 18.4 megabytes per second > Add the fact that X server does just a memcpy in the worst case (and exactly nothing in the best case), I don't know anybody who still uses a VGA 640x480 display with 16-bit color depth - my display is 1920x1280x4 * 30fps = ~280mb/s at 30 fps. Furthermore, X has to move the data over busses to vram and in the worst case do a blit using the CPU. >From my personal experience, compute-bound workload often tends to be easier to handle than bandwidth-bound stuff where suddenly bottlenecks are waiting all along the data paths (I do performance-tuning for VLIW DSP chips as a living). Anyway, in my profile I don't see any resampling. Either I am really hitting that bug, or it is another inefficiency happening somewhere. Regards, Clemens
