Page 1 of 1

Linux power regression problem

Posted: Sat Nov 12, 2011 12:25 pm
by xircon
Interesting article here:
http://www.omgubuntu.co.uk/2011/11/linu ... pad-fixed/

I am testing it now, definitely appears to slightly lower the temperature.

Steve

Re: Linux power regression problem

Posted: Fri Nov 18, 2011 11:20 am
by ej64
on my X220 the Debian 2.6.39 kernel with pcie_aspm=force works best, idling at ~ 6W.

Damentz has patched his latest liquorix kernel (3.1.0-1.dmz.2-liquorix) accordingly but this strangely helps nothing with my machine -> Liquorix forum.

Re: Linux power regression problem

Posted: Tue Nov 22, 2011 8:48 am
by ej64
An update:

Since kernel 3.0 the behavior of the i915 (new sandy bridge chipset) power states have been altered. By default the low power states are disabled due to bug reports on some machines. You can revert this power regression with boot parameter "i915.i915_enable_rc6=1" (check, if anything still works fine!). On my machine this lead to a reduced display resolution of 1024x768 making it necessary to use an additional "vga=844".

Together with Damentz' latest Liquorix kernel (3.1.0-1.dmz.2-liquorix-amd64) that is patched with Matthew Garrett's patch series I now get the lowest power consumption ever on my machine (~ 5.4 W). :D

full story: Liquorix forum

Re: Linux power regression problem

Posted: Tue Nov 22, 2011 10:34 am
by xircon
How do you measure your power consumption?

Not had much luck with Liquorix before, could never get FGLRX to load, but being as FGLRX is trashed anyway, will give it another go.

Re: Linux power regression problem

Posted: Tue Nov 22, 2011 12:07 pm
by ej64
xircon wrote:How do you measure your power consumption?
Usually there's a directory BAT0 deep in the tree /sys or /proc with some statistical files (e.g. look for /proc/acpi/battery/BAT0/state). In my case it's an infrastructure built by tp_smapi (Thinkpad specific) giving lots of infos, one is /sys/devices/platform/smapi/BAT0/power_now. I think it's reliable.
Not had much luck with Liquorix before, could never get FGLRX to load, but being as FGLRX is trashed anyway, will give it another go.
Liquorix usually was fine to me, but every kernel update shoots your fglrx to pieces. That sucks (previous I had an Asus with AMD cpu and integrated HD3200). If you're on testing or sid you should be aware of possible incompatibilities between fglrx and xorg-server, e.g. only the latest Catalyst 11.11 supports xorg-server 1.11.
If your fglrx is screwed up, you should try to uninstall it (/etc/ati/fglrx-uninstall.sh) and/or use kanotix' or sgfxi script to install (only from console! X not running!!). Man, am I happy that I got rid of it ... :mrgreen:

Re: Linux power regression problem

Posted: Tue Nov 22, 2011 12:54 pm
by xircon
Yeh, just tried the FGLRX from experimental, doesn't work at all well in Gnome3, have dropped back to the Gallium drivers.
Battery info:

Code: Select all

cat state 
present:                 yes
capacity state:          ok
charging state:          charged
present rate:            1 mA
remaining capacity:      4500 mAh
present voltage:         12326 mV
molly2@n5010 /proc/acpi/battery/BAT0 $ cat info
present:                 yes
design capacity:         4500 mAh
last full capacity:      4433 mAh
battery technology:      rechargeable
design voltage:          10800 mV
design capacity warning: 450 mAh
design capacity low:     136 mAh
cycle count:		  0
capacity granularity 1:  45 mAh
capacity granularity 2:  45 mAh
model number:            DELL 4YRJH03
serial number:            1151
battery type:            LION
OEM info:                Sanyo
Can't see anything useful.

Re: Linux power regression problem

Posted: Tue Nov 22, 2011 1:39 pm
by ej64

Code: Select all

cat state 
present:                 yes
capacity state:          ok
charging state:          charged
present rate:            1 mA
remaining capacity:      4500 mAh
present voltage:         12326 mV
Can't see anything useful.
Well, you will see nothing useful furthermore if it's not on battery ... :mrgreen:
And some physics fundamentals would be good :wink: W = V*A

In your case you can't see Watts in the first place, the battery voltage will change over time, therefore the present rate figure (mA = milliAmpere = 0.001 Ampere) will change while discharging, too, giving you no reliable number. You have to multiply present rate and present voltage to get the actual Watts (note: 12326 mV = 12.326 V)

Got it? :)

Re: Linux power regression problem

Posted: Tue Nov 22, 2011 2:00 pm
by xircon
Long time since I needed any of the Physics I learned at school (approx 35 years ago), the earth was flat then :lol: