On 04/08/2011 12:11 AM, Tom Gundersen wrote:

On Fri, Apr 8, 2011 at 10:38 AM, Matthew Monaco wrote:

...

I like having volume control for HDMI out. And I like that this mysterious starving audio, playing silence, glitch has disappeared. But I don't like that pulse uses so much CPU and i think it's a major reason why my computer is runn 5-8 degrees C hotter with gnome3 FWIW, this is almost certainly a bug that can be fixed. In my experience PA allows huge power savings compared to ALSA (down from more than 100 wakeups/sec to less than 10 wakeups/sec).

Try debugging a bit to see what is going on powertop2 should be very helpful.

Cheers,

Tom It could also be that GNOME 3 uses hardware acceleration by default[1]. If weren't running Compiz before, it might be making a difference (I have no idea how it compares if you were using Compiz before).

[1] http://www.gnome3.org/faq.html - See "Will my computer be able to run GNOME 3?"

On 08/04/11 22:54, Bernardo Barros wrote:

Hi guys,

Can't pulseaudio be moved to optdepends?

No. It is an upstream requirement.

On Sat, Apr 9, 2011 at 8:21 AM, Matthew Monaco wrote:

On 04/08/2011 02:11 AM, Tom Gundersen wrote:

...

On Fri, Apr 8, 2011 at 10:38 AM, Matthew Monaco  wrote:

...

I like having volume control for HDMI out. And I like that this mysterious starving audio, playing silence, glitch has disappeared. But I don't like that pulse uses so much CPU and i think it's a major reason why my computer is runn 5-8 degrees C hotter with gnome3

FWIW, this is almost certainly a bug that can be fixed. In my experience PA allows huge power savings compared to ALSA (down from more than 100 wakeups/sec to less than 10 wakeups/sec).

Try debugging a bit to see what is going on powertop2 should be very helpful.

Cheers,

Tom

I don't see how your CPU can be _less_ active with PA. I thought PA added a layer that didn't exist before and didn't actually remove anything.

I'll say though, that changing VLC from 'default' or explicitly ALSA to PulseAudio dropped CPU usage of PA from ~16% to 3%. (It's mysterious to me why that 3% remains when the media is paused though).

I guess I need to go through gstreamer and other app-specific settings and explicitly point to PA.

PA buffers better than ALSA, or is supposed to, in any case. Of course, if you're using its ALSA-emulation that's a moot point. AFAICR (haven't had this problem for ages), CPU usage from PA is basically caused by the resampling, which you can just modify to a simpler method if you don't need the higher quality of the default (speex, I believe). So yes, PA would save CPU cycles and battery (its meant for use on cellphones as well, after all, the nokia people were involved at some point).

On 04/09/2011 04:46 AM, Mauro Santos wrote:

On 09-04-2011 02:12, Oon-Ee Ng wrote:

...

PA buffers better than ALSA, or is supposed to, in any case. Of course, if you're using its ALSA-emulation that's a moot point. AFAICR (haven't had this problem for ages), CPU usage from PA is basically caused by the resampling, which you can just modify to a simpler method if you don't need the higher quality of the default (speex, I believe).

I have been using the ffmpeg resample method because of that, it uses less cpu, or seems to use less cpu here, and with my setup I cannot detect any loss in quality, but hey, it is a laptop and the analog audio path is not stellar so ymmv.

Hmm, I tried that, didn't seem to help. In fact it's super frustrating now because VLC is back to use ~16% CPU again no matter what resample method I use and what audio output option I choose. Also, does anyone know what is restarting pulseaudio after I --kill? Is it gnome-shell?

On Sat, Apr 9, 2011 at 9:24 AM, Sander Jansen wrote:

On Sat, Apr 9, 2011 at 8:47 AM, Matthew Monaco wrote:

...

On 04/09/2011 04:46 AM, Mauro Santos wrote:

...

On 09-04-2011 02:12, Oon-Ee Ng wrote:

...

PA buffers better than ALSA, or is supposed to, in any case. Of course, if you're using its ALSA-emulation that's a moot point. AFAICR (haven't had this problem for ages), CPU usage from PA is basically caused by the resampling, which you can just modify to a simpler method if you don't need the higher quality of the default (speex, I believe).

I have been using the ffmpeg resample method because of that, it uses less cpu, or seems to use less cpu here, and with my setup I cannot detect any loss in quality, but hey, it is a laptop and the analog audio path is not stellar so ymmv.

Hmm, I tried that, didn't seem to help. In fact it's super frustrating now because VLC is back to use ~16% CPU again no matter what resample method I use and what audio output option I choose.

Also, does anyone know what is restarting pulseaudio after I --kill? Is it gnome-shell?

I believe it uses dbus activation so it could be any pulseaudio client...

Actually, not sure if that's entirely true what I just said. Now '/etc/pulse/client.conf' does have a autospawn configuration option which might prevent it from restarting.

On 09-04-2011 14:47, Matthew Monaco wrote:

Hmm, I tried that, didn't seem to help. In fact it's super frustrating now because VLC is back to use ~16% CPU again no matter what resample method I use and what audio output option I choose.

Are you sure you have set vlc to output to pulse directly? It works fine here with minimal cpu usage. Also make sure you dont have any pavucontrol or volume metering apps running. I've found those can be quite taxing on the cpu. -- Mauro Santos