Kernel Mode Setting + Dynamic Clocks

[agd5f]

The flicker is due to changing the clock during scanout. The driver tries to change the clock during the vblank period, but it doesn't quite work right yet.

[DF5JT]

yeah, the flickering is really annoying - but I have already seen a patch for this on the DRM mailing list. Did not try it out yet though...

I am a little confused about all the various branches flying around. The dynpm=1 works with 2.6.33 from airlied's drm-next-radeon git repository, however, no new patches have been added there to get rid of the flickering. I am willing to test anything, so please point me to something that I can help with. Ihave a FireGL V5200 in my notebook.

as for always forcing the lowest power mode: that might not result in the most power savings. see the CPU situation
http://www.mjmwired.net/kernel/Docum.../governors.txt

I thought that keeping core voltage and frequency at one constant level would get rid of the annoying flicker.

[madman2k]

Well, I'd like to see an option that statically sets the lowest possible energy usage for notebook operation.

Having said that, already now the dynpm=1 dramatically improved thermal performance of my T60p. Were it not for the annoying flicker from changed power states, this new development really made fanless operation possible. In idle mode the GPU settles at 71/72?C which is acceptable, since it is well within the specs of the chip.

I am pleased and very much hope that this flickering will be eliminated soon.

yeah, the flickering is really annoying - but I have already seen a patch for this on the DRM mailing list. Did not try it out yet though...

as for always forcing the lowest power mode: that might not result in the most power savings. see the CPU situation
http://www.mjmwired.net/kernel/Docum.../governors.txt

[DF5JT]

well actually it already does memory and core reclocking as well as reducing the number of PCIE lanes - the only thing I miss now is core downvolting...

Well, I'd like to see an option that statically sets the lowest possible energy usage for notebook operation.

Having said that, already now the dynpm=1 dramatically improved thermal performance of my T60p. Were it not for the annoying flicker from changed power states, this new development really made fanless operation possible. In idle mode the GPU settles at 71/72?C which is acceptable, since it is well within the specs of the chip.

I am pleased and very much hope that this flickering will be eliminated soon.

[dalingrin]

Probably you have a mix of distribution released components and bleeding edge stuff. Yesterday I went through the whole process of installing drm, mesa and xf86-drivers-ati as well as airlied's drm-next-testing stuff from git and I managed to get dynamic power management running with KMS on a FireGL V5200 (r300). PM ist still in rudimentary form but development is progressing quite nicely and I expect to update all the stuff by next week and have the biggest annoyance (horizontal flicker when power states change) gone by then.

It is very helpful to join IRC channels #dri-devel and #radeon as well as reading the dri-devel mailing list. It is quite a complex issue.

HTH

I should have mentioned that I already have mesa, xf86-video-ati, libdrm, etc all compiled from git.

Thanks for the pointer to the mailing list and IRC.

[madman2k]

PM ist still in rudimentary form

well actually it already does memory and core reclocking as well as reducing the number of PCIE lanes - the only thing I miss now is core downvolting...

[DF5JT]

I'm a totally a newb at the ATI open source landscape.
I currently have the latest drm-radeon-testing installed with 2.6.33. I am trying to use kms with dynamic clocks on a Mobility Radeon 5830.

KMS seems to work fine but as soon as I boot with radeon.dynpm=1, I get a black screen and an apparent lock.

I don't see any relevant errors in any logs.

Is this expected behavior because dynamic clocks is not implemented with evergreen or, is there something else going on?

Probably you have a mix of distribution released components and bleeding edge stuff. Yesterday I went through the whole process of installing drm, mesa and xf86-drivers-ati as well as airlied's drm-next-testing stuff from git and I managed to get dynamic power management running with KMS on a FireGL V5200 (r300). PM ist still in rudimentary form but development is progressing quite nicely and I expect to update all the stuff by next week and have the biggest annoyance (horizontal flicker when power states change) gone by then.

It is very helpful to join IRC channels #dri-devel and #radeon as well as reading the dri-devel mailing list. It is quite a complex issue.

HTH

[Melcar]

Not with KMS. Radeon alone has DynamicPM, ClockGatting, and ForceLowPowerMode, all of which reduce core clocks in varying degrees. I think there is some initial work being done with KMS power management, but I haven't been able to test it out yet.

[dalingrin]

There is no power management with KMS yet, I don't think.

I was under the impression that drm-radeon-testing had some new dynamic clock algorithms.

[Melcar]

There is no power management with KMS yet, I don't think.