Bug 31588 - [SNB] Idle power consumption on Huron River is too high: 11+w vs 8w
Summary: [SNB] Idle power consumption on Huron River is too high: 11+w vs 8w
Status: CLOSED FIXED
Alias: None
Product: DRI
Classification: Unclassified
Component: DRM/Intel (show other bugs)
Version: unspecified
Hardware: Other All
: medium critical
Assignee: Chris Wilson
QA Contact:
URL:
Whiteboard:
Keywords:
Depends on:
Blocks:
 
Reported: 2010-11-12 12:04 UTC by Manjeet Singh
Modified: 2017-07-24 23:06 UTC (History)
2 users (show)

See Also:
i915 platform:
i915 features:


Attachments

Description Manjeet Singh 2010-11-12 12:04:40 UTC
Discription- We did the power managment on latest our latest SI-2 Huron river plateform. The power numbers we are getting idle power numbers always 11+ watts in HP Quickweb (linux based invironment).

Hardware: Huron River UMA
OS- HP Quickweb (fedora based linux).
Kernel- 2.6.35-ng #1
X driver- xf86-video-intel 2.12.

Steps:
Boot HP- SI-2 unit on battery
No external device connected
Wifi not associated
Brightness is set to 50%
Leave the system in complete idle state (absolutly no activity)
Measure the power consuption using powertop.

The idle Power consumption is coming 11+ watts in HP Quickweb.
In windows the idle numbers are better ~8 watts.

We believe that Intel linux driver does not idle as efficiate as windows driver does.
Comment 1 Chris Wilson 2010-11-22 03:39:08 UTC
One of the baselines that would be interesting is the idle power without i915.ko loaded. In theory, loading i915.ko should enable power saving...
Comment 2 Manjeet Singh 2010-12-14 11:34:33 UTC
Hello,

The current driver does not contain any support for setting up the watermark registers in SandyBridge

We also run many experiment and compared with windows. See below. 

Linux and windows numbers are quite close when Display is off with backlight set to ZERO. But when brightness is set to 50% or 100% the differences between quick and windows become ~3 watts. Looks like linux driver is not handling the graphics power part the way windows driver does:



Test using powertop and batterymon   Intel UMA: Quickweb	Windows 7
Idle power  (50% brightness)	                  11 W	       7.6 W
Brightness is 100%,	                          12.7 W	9.8 W
Brightness is  0%  	                          10.6 W	7.5 W
Backlight is Off 	                          7.7 W	    7.2 W (Dim Display)
Display off	                                  7.7 W	        6.5 W
Comment 3 Chris Wilson 2010-12-15 05:24:29 UTC
Manjeet, thanks for the testing. Those numbers look really useful. On drm-intel-next, we're experimenting with enabling RC6, CxSR and FBC, which should reduce our power consumption, especially the 3W disparity on enabling the pipe.
Comment 4 Gordon Jin 2010-12-15 17:19:37 UTC
that won't be in .37 kernel.
unblocking Q4 release.
Comment 5 Yuanhan Liu 2010-12-16 02:24:55 UTC
Hi,

Here is the result of my roughly test which is Just For Your Information.

Sorry, I have no windows 7 installed in that test machine.

Machine: Dell 
kernel: 2.6.37-rc5, with Self-refresh and FBC enabled, which can save about 1W.


Brightness 100%:            13.4 W
Brightness 50% :            11.5 W
Brightness 0%  :            9.9  W
Display off    :            8.5  W


Thanks.
Yuanhan Liu
Comment 6 Manjeet Singh 2010-12-17 15:08:29 UTC
Yuanhan’s numbers pretty much sync up with our QW numbers

Question:

-Is this with the latest code that Chris points to?  If so there aren’t any savings?
Comment 7 Michael Fu 2011-01-05 20:26:40 UTC
In reply to comment #3)
> Manjeet, thanks for the testing. Those numbers look really useful. On
> drm-intel-next, we're experimenting with enabling RC6, CxSR and FBC, which
> should reduce our power consumption, especially the 3W disparity on enabling
> the pipe.

RC6, CxSR and FBC are in -next now. Those are pretty much the idle power feature we can utilize in linux on SNB. Mark this bug as fixed.

BTW, the test result can only be used for reference - gfx driver alone can't fix the power delta between windows and linux.


Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.