6 Replies Latest reply on Sep 22, 2013 3:52 PM by blu

    Unexpected drop in GPU core clock on C60 APU

    blu

      Hi,

       

      I've been using two C-series APUs in tandem for benchmarking and unit-testing of proprietary OGL/OCL software. The APUs are running Ubuntu Oneiric installations, which even though not identical, have been kept largely in-sync for the purposes of development - same kernel versions, same Catalyst/fglrx versions, same compiler toolchains, etc. Now, one of the APUs is a C-60, and the other is a C-70. Everything has been going smoothly until a few days ago when the C-60 started exhibiting an abnormal performance degradation - all my benchmarks started showing a performance gap between the two units. Poking around I noticed that the GPU core clock of the underperforming unit (C-60) was below its nominal value:

       

      $ aticonfig --init --odgc
      Uninitialised file found, configuring.
      
      Adapter 0 - AMD Radeon HD 6290 Graphics
                                  Core (MHz)    Memory (MHz)
                 Current Clocks :    200           533
                   Current Peak :    275           533
        Configurable Peak Range : [275-275]     [533-533]
                       GPU load :    0%
      $ aticonfig --init --pplib-cmd "get clock"
      Uninitialised file found, configuring.
      Engine Clock Range: 200-200 MHZ; Memory Clock Range: 533-533 MHZ;

       

      In comparison, its sibling C-70 has been running all normal (both systems are running on AC power):

       

      $ aticonfig --init --odgc
      Uninitialised file found, configuring.
      
      Adapter 0 - AMD Radeon HD 7290 Graphics
                                  Core (MHz)    Memory (MHz)
                 Current Clocks :    275           533
                   Current Peak :    275           533
        Configurable Peak Range : [275-275]     [533-533]
                       GPU load :    0%
      $ aticonfig --init --pplib-cmd "get clock"
      Uninitialised file found, configuring.
      Engine Clock Range: 275-275 MHZ; Memory Clock Range: 533-533 MHZ;
      

       

      (the recurrent use of --init flag and corresponding 'configuring' message is due to the absence of xorg.conf on either system)

       

      My efforts to get the core clock of the 'slacking' unit back to 275MHz have been fruitless. Perhaps because I have no idea what caused the drop in the clock in the first place. Any ideas for getting the 'ailing' unit back to full speed, or what could have causes the drop, would be greatly appreciated.

       

      Regards,

      Martin

        • Re: Unexpected drop in GPU core clock on C60 APU
          hsaigol

          Hi Martin,
          Can you list the steps that you have already tried?
          I'm guessing you've tried all of the following
          1) Re-installing the driver

          2) Resetting the overclock profile

          3) Check the temperature of the C-60 APU (ensure it isn't overheating)

          4) Ensuring no updates (possibly automatic) were installed between the time frame of working ok and the issue

          5) Can you check the system BIOS of the C-60 system and see if the power adapter is correctly recognized (if it supports that feature). I have seen sometimes the charger will have issues and the APU will revert back to DC power state thus lower and immovable clocks. If the adapter is compatible between your C-60 and C-70 system please try to swap them and do a full power cycle on the C-60 system.

           

          Anything you can remember in terms of changing system config/drivers before you started seeing the issue?

          1 of 1 people found this helpful
            • Re: Unexpected drop in GPU core clock on C60 APU
              blu

              Hi hsaigol,

               

              I have tried the following:

              1) Re-installed Catalyst12.8 through 13.4 repeatedly (my first reaction). Currently at 13.4, but that does not make any difference for the issue at hand.

              2) Issued --odrd, --od-disable, --od-enable; uninstalled Jupiter (a performance manager applet; just in case, even though it runs happily on the C-70 unit). It bears noting that --odsc 275,533 on that machine always fails with an 'ERROR - ATI Overdrive (TM) is not supported on Adapter 0 - AMD Radeon HD 6290 Graphics', but I guess that's normal given how the configurable peak range is one value.

              3) The temperature is normal - at least not higher than what I've observed before the drop - gets to high 70s/low 80s at peak loads. At no load it's in the high 50s.

              4) Automatic updates are disbaled on both machines.

              5) I have not been able to find power adapter info in the BIOS, but Xorg.log says 'fglrx(0): AC Adapter is used', so I guess the power detection works normally.


              Now, re the events preceeding the issue, I did try the new beta (13.8 beta2) a week-or-so ago on that machine, but reverted to the official version effectively immediately. Unfortunately I don't remember if that produced any performance delta. i.e. if the issue had crept up before that.

            • Re: Unexpected drop in GPU core clock on C60 APU
              blu

              Good news: yet another driver re-install has solved the issue. I still have no idea what caused it in the first place (and why it persisted across multiple driver re-installs), but since the core clock is back to normal now I'm tagging the subject as solved. Credit goes to hsaigol for the helpul comment.