More Radeon Power Management Improvements

[yotambien]

May someone post his experiences with power management in OSS driver in comparison to Catalyst? In my case i lost like 33% battery life :/

With the addition of power profiles I'm very happy. I basically can do what I used to with Catalyst, this is, set the lower power profile and forget about it. The comparison is not totally fair though, for the performance with fglrx in the low profile (any profile, really) was much better than with radeon.

I haven't timed how long the battery lasts in comparison, but I believe there's still some work to do with radeon. At least, according to the feature table in freedesktop, it seems that the memory reclocking thingy is not done yet.

[Rabauke]

It would be kind if someone could edit http://www.x.org/wiki/RadeonFeature and add a short synopsis of how en-/disabling radeon power management and what options further are available and how to read out temperatures. I think, there is something semi-official like this missing.

Alex, you are awesome! Many thanks for this. So, for power management questions, this site is now a nice place to go! Just one tiny little remark: Please specify the kernel version which is at least needed to benefit from pm.

Cheers

[HokTar]

In case anybody is interested: these are the steps for Ubuntu and many others to set a power profile:

Check your current state:
cat /sys/kernel/debug/dri/0/radeon_pm_info

Use dynpm:
sudo -i
echo dynpm > /sys/class/drm/card0/device/power_method
exit

Please note that Alex wrote card-0 but in Ubuntu (and possibly others) it is card0!

Use a profile:
sudo -i
echo profile > /sys/class/drm/card0/device/power_method
echo low > /sys/class/drm/card0/device/power_profile
exit

Of course you can use
-default
-auto
-high
-mid
-low
instead of low.

Now you can recheck if your power state changed to the desired one:
cat /sys/kernel/debug/dri/0/radeon_pm_info

I hope it may make life easier for some.

[Rabauke]

While setting power_methods and _profiles works fine, I'm still having trouble with reading out GPU temperature. My card is a rv710 0x954f. sysfs gives me three hwmon entries: acpitz-virtual-0; atk0110-acpi-0; k8temp-pci-00c3. sensors-dedect from lm-sensors doesn't give me more. Is it possible, my card fails to have a thermal sensor? I'm running 2.6.35-trunk.

[agd5f]

Check the dmesg output. If your card has a thermal sensor (not all OEMs implement it), it should say something like:
[drm] Internal thermal controller with fan control
and the name of the sensor in sysfs should be radeon:
cat /sys/class/hwmon/hwmon0/name
radeon

[Rabauke]

[ 5.220776] [drm] Internal thermal controller with fan control
But none of the hwmon's is related to radeon. If I got the above message, do I have one, or is it still unclear?

[agd5f]

internal thermal sensor support was added in 2.6.36, so you need a newer kernel.

[Rabauke]

Ah, the piece of information I missed. Thanks again. Your work and feedback is really appreciated!

[Panix]

Can I interrupt to ask what the current status is for power management is for Evergreen cards?

For e.g., say, a HD 5750 card? is there power management profiles or support for such cards? IIRC, there was discussion on this before and it was 'in development' or 'on the way' or something. I'm just wondering what the progress has been since then. If it's not there, is there any sort of time schedule for it? I suppose, if it's not working, the user will probably choose to install Catalyst drivers until it's working???

I appreciate any answers or replies to that inquiry... sorry to interrupt...

I suppose the power management inquiry is most applicable or of interest to those with AMD/ATI GPU (discrete card) laptops?

[bridgman]

I think it's the same as for previous generations - choice between dynpm or profile-based, and if you chose profile-based there is a choice of low/med/high/auto profiles. Dynpm is glitchy on some systems and probably needs more work to resolve that (ie more than "bug fix").