X.Org ATI Driver Supports New Power Options

[bridgman]

It almost seems to me that my GPU is always under load with the radeon driver, since while assumedly idling it goes up to temperatures that fglrx reached only under load. Time for a bug report maybe?

The fglrx driver has kernel code that dynamically adjusts clock frequencies based on 2D and 3D load. The open drivers are still doing power management in the user-space driver so they can't do load-based clock switching yet. Before Alex added the recent power management options the open drivers ran the chip at full speed, while fglrx was downclocking it most of the time to save power. Using the ForceLowPowerMode option slows down the engine clock and should make the power consumption and heat generation more like fglrx.

Once the kms/mm and 6xx/7xx 3D work in the kernel driver (drm) settles down and the APIs stabilize I imagine you'll see power management code move into the kernel and become more "dynamic". Anyways, not a bug, just lack of an architectural feature.

[oibaf]

Or apply this patch bridgman talked about and alex created. Very nice one:

http://www.botchco.com/alex/xorg/print_clock_pm.diff

Why not applying this patch upstream?

[bridgman]

The patch adds log entries on every clock change, so the log file would grow without limit. I don't know "how many is too many" but continuously logging events is generally frowned on unless you have a logging system which only keeps the last N entries.

[oibaf]

The patch adds log entries on every clock change, so the log file would grow without limit. I don't know "how many is too many" but continuously logging events is generally frowned on unless you have a logging system which only keeps the last N entries.

Looking at the patch it appears that it only adds the printed clock value, while before it only printed the switch, so no new entries are printed:

current:
Power Mode Switch

with patch:
Power Mode Switch: %d Mhz

[agd5f]

The info is there for debugging purposes at the moment. At some point, we'll get rid of it or make it a debug option.

[bugmenot]

As long as it is for debugging purposes I think it would make sense to show the mhz. I'm also for apply this patch upstream. We linux user want to know whats going on.

[Ant P.]

Just installed latest xorg git to try this out. The radeon driver hasn't crashed yet, but as a side effect I have a horribly broken evdev keyboard driver now. *sigh* :-/

(edit: never mind - that was just my xmodmap stuff screwing things up)

[planetf1]

I've just tried enabling this on a Thinkpad T60p, but see the following in Xorg.0.log

(WW) RADEON(0): Option "ClockGating" is not used
(WW) RADEON(0): Option "ForceLowPowerMode" is not used
(WW) RADEON(0): Option "DynamicPM" is not used

I have this in the device section in xorg.conf
#Option "DynamicClocks" "on"
Option "ClockGating" "on"
Option "ForceLowPowerMode" "on"
Option "DynamicPM" "on"

SW levels are from Fedora 11
xorg-x11-drv-ati-6.12.2-15.fc11.i586
kernel-PAE-2.6.29.4-167.fc11.i686

From xorg log:
(II) Module radeon: vendor="X.Org Foundation"
compiled for 1.6.1.901, module version = 6.12.2
Module class: X.Org Video Driver
ABI class: X.Org Video Driver, version 5.0
(--) RADEON(0): Chipset: "ATI Mobility FireGL V5200" (ChipID = 0x71c4)

Why aren't these options effective, what am I missing?

Am I mistaken in thinking these changes made it into the fc11 rpm?

Is there also any way to check the current clock freq?

Thanks

[Zhick]

@planetf1: The powermanagement-options aren't yet in a release, so you need the git-master version of xf86-video-ati. 6.12.2 is too old.

[planetf1]

Thanks. I've not built ati from git before, but can give it a go.

One thought though - with a number of dependencies between kernel/kms/libdrm/mesa, are there any particular recommendations on building a suitable git version for F11?

At this point I'm wondering if it just makes sense to wait until the maintainer picks this up.