No announcement yet.

Power question

  • Filter
  • Time
  • Show
Clear All
new posts

  • Power question

    mmm could be possible to add a GPU temp/load/freqs monitor in the OSS driver?

    i dont mean a full catalyst overdrive stuff i just mean like some dump info in proc for example like /proc/cpuinfo.

    i ask cuz well is not possible to know how hot is my gpu or which frequency is actually using while using kwin composite for example or even how much load the gpu really is. so as far as i know just running X server in my PC could eventually burn my gpu without even notice it.

    it worries me not to know cuz well in linux my 4850x2 is always at 100% fan speed (i know cuz it sounds like a 747 at full throtle landing), so how i know why the fan is always 100% without matter any xorg.conf paramater i put or take, is fine be always 100%? kms is hardfixed to 100% just in case until pms comes?

  • #2
    I have also been waiting for this, but my guess is that until proper power management is implemented we won't see anything like this. Radeonhd does report clocks tough (and it also reads all the available frequency steps in the card's BIOS).


    • #3
      There's some preliminary dynamic clocking code in the kernel, not sure if it's in mainline but in drm-radeon-testing there is. You should be able to read back the clocks with something like this:

      cat /sys/kernel/debug/dri/0/radeon_pm_info

      If the dynamic clocking code isn't working for you it's very trivial to manually down clock your card, just add calls to these functions to the driver.

      Works like a charm here.


      • #4
        The work agd5f has been doing on the I2C support is the first step, at least for most of the chips/cards out there. Once the I2C support is stable then the third party fan/thermal controller chips can be queried to see what the temps are like.

        The I2C support in turn depended on KMS being stable; you can kinda see the sequence that needs to play out here...


        • #5
          For clarity, our understanding is some of the newer chips don't need to go through I2C for this info since more of the logic is on chip -- for those chips it's more a question of digging up the info, determining what can be released, and writing up code and/or documentation.