Leaving the Desktop Market

Alexey Dokuchaev danfe at nsu.ru
Thu Apr 3 07:01:36 UTC 2014


On Wed, Apr 02, 2014 at 11:39:07PM -0700, Kevin Oberman wrote:
> On Wed, Apr 2, 2014 at 7:14 PM, Alexey Dokuchaev <danfe at nsu.ru> wrote:
> > [...] The setup is a bit ugly, but I only had to check the available PINs
> > (ugly, ugly) and set up stuff once. It just works.
> >
> > Not always, unfortunately.  I also had a working pin override configuration
> > in /boot/loader.conf, but after r236750 (major snd_hda driver rewrite) it
> > stopped working.  I've reported it and tried to get some support from mav@
> > but he never replied.  Since then, I have to carry pre-r236750 version of
> > snd_hda(4) to have working sound.
> 
> Is that just in head? Do I have more fun to look forward to?

r236750 is MFC to stable/8 (yeah, it went pretty far; if I had been doing
upgrades more often I would probably have caught it and vetoed MFC, but we
are now here where we are, so I guess I have to live with it until I find
time to sit down and figure out what went wrong with my setup).  That said,
if everything keeps working for you, then you probably should not worry. :)

> The key problem with power, as I have written several times is the
> conflation of TCC or throttling as power management tools. Mix them (they
> really don't save power) with Cx states is often worse than what you are
> seeing. It can cause many systems to lock up.
> 
> Try setting:
> powerd_enable="YES"
> performance_cx_lowest="Cmax"
> economy_cx_lowest="Cmax"
> into /etc/rc.conf and putting:
> # Disable CPU throttling
> hint.p4tcc.0.disabled=1
> hint.acpi_throttle.0.disabled=1
> into /boot/loader.conf. That should work MUCH better and will really save
> power (assuming that your system supports better than C2 as C2 usually is
> a pretty minor power savings. C3 or higher is usually where things really
> start to improve.

Wow, that's great, thanks for your advice!  I'll try these out and see how
it will go.

> I've read a paper from SDSC (San Diego Supercomputer Center) showing that
> CX states are by far and away the most significant power saver and they
> should cause only very trivial and unnoticeable impact on performance.
> Number two is EST, but that is almost always enabled on FreeBSD, so I
> assume that you have that running already (or the AMD equivalent).

Noted.  FWIW, I do have EST enabled; I'm also running it with modified DSDT
file (patched _PSS table) for CPU undervolting (less heat, longer battery
life).

./danfe


More information about the freebsd-hackers mailing list