Announcement

Collapse
No announcement yet.

NVIDIA Recalls Linux Drivers Over Fan Speed Bug

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • NVIDIA Recalls Linux Drivers Over Fan Speed Bug

    Phoronix: NVIDIA Recalls Linux Drivers Over Fan Speed Bug

    This week NVIDIA had to pull its latest WHQL-certified graphics drivers on Windows due to a bug that would cause the fan controller to not respond correctly to the current conditions of the GPU workload and in some cases would even turn the GPU's fan off. This bug could potentially kill the NVIDIA graphics card due to overheating...

    http://www.phoronix.com/vr.php?view=ODA0MA

  • #2
    Ahh well at least it's WHQL on Windows :P

    Comment


    • #3
      Don't drivers have some sort of shut off mechanism that prevents damage to the GPU due to overheating? If not then they should.

      Comment


      • #4
        Just one more reason to not trust binary blobs.

        Comment


        • #5
          Originally posted by Melcar View Post
          Don't drivers have some sort of shut off mechanism that prevents damage to the GPU due to overheating? If not then they should.
          I don't think drivers are really the right place for that sort of thing. It ought to be impossible (or nearly so) for the host to fry the GPU even with malicious driver code.

          Comment


          • #6
            I'd say the GPU bios should contains some kind of self protective killswitch code that kills the card if the temperature goes too high.

            Comment


            • #7
              it makes me wonder how much is that WHQL certification worth, if such a severe problem passed unnoticed.

              Comment


              • #8
                It's just money really. I doubt there is any strict quality control going on with certification. You're just paying for the privilege of your software to be able to run on Windows OS. I doubt the "certification" means anymore than "does it run on our OS?".

                Comment


                • #9
                  I am not a lawyer, but I suspect that if WHQL certified drivers killed my GPU, I would have a case against Microsoft, even if I don't have a case against the GPU vendor. The second line of the Windows Logo Program page [1] reads:
                  The Windows logo signifies the compatibility and reliability of systems and devices with Windows operating system.
                  I fail to see how a driver that kills my card can be advertised as "compatible and reliable", which means that Microsoft could be sued under the product description laws, not to mention "fitness for purpose"! After all, if it's WHQL, it's Microsoft who are asserting the compatibility, not the GPU vendor.

                  That said, I'm not sure anyone has ever tried to take a case like that, so it's likely to be a very interesting one if it happened.

                  [1] - http://www.microsoft.com/whdc/winlogo/default.mspx

                  Comment


                  • #10
                    Watercooling FTW

                    Comment


                    • #11
                      Originally posted by Max Spain View Post
                      Watercooling FTW
                      Passive cooling for the win...

                      Comment


                      • #12
                        Basically they are paying 250$ per test just to be able to put a windows logo sticker on their product. Read up some stuff on wikipedia and they seem to run some basic tests on the driver, wouldn't be surprised if it was fully automated and would take a hour max without any kind of input.

                        But this case just proves the drawbacks of em old binary blobs. Too bad Nvidia isn't planning on going open source or at least letting more people in on the thing.

                        Comment


                        • #13
                          I don't see how this is a problem exclusively related to closed source drivers.

                          Comment


                          • #14
                            Originally posted by Melcar View Post
                            Don't drivers have some sort of shut off mechanism that prevents damage to the GPU due to overheating? If not then they should.
                            they do.
                            and that's what happens here; the driver fails to act on a soft limit and the hardware shuts down due to a hard limit.
                            You should be able to use the card again after it has cooled down (though doing this constantly will probably damage the card eventually)

                            Comment


                            • #15
                              Originally posted by Max Spain View Post
                              Watercooling FTW
                              jup. When you see smoke, just dump a bucket of water on it

                              Comment

                              Working...
                              X