CUDA on Linux

Forums › Cruncher's Corner

microchip
microchip
Joined: 10 Jun 06
Posts: 50
Credit: 218927781
RAC: 193979
15 Apr 2011 16:57:25 UTC
Topic 195753
(moderation:
    )

    Hi,

    Got a question here. I'm fairly new to CUDA and would like to add a CUDA capable GPU in my crunch system. I'm on Linux and my question is if I have to install anything special to make it work with BOINC and E@H? I mean do I need the proprietary driver from nVIDIA and do I need some specific CUDA kit for it to work?

    Thanks :)

    Michael Karlinsky
    Michael Karlinsky
    Joined: 22 Jan 05
    Posts: 888
    Credit: 23502182
    RAC: 0

    CUDA on Linux

    15 Apr 2011 17:18:30 UTC
    Message 105095
    (moderation:
      )

      Hi,

      you need to install the NVIDIA driver.

      download

      and

      instructions

      excerpt:

      1) Kernel sources must be installed and configured. Usually this means
installing the 'kernel-source', 'make' and 'gcc' packages with YaST2.

      2) Run the following commands

      echo "blacklist nouveau" > /etc/modprobe.d/nvidia.conf
      # recreate initrd without KMS, if the use of KMS is enabled in initrd
      if grep -q NO_KMS_IN_INITRD=\"no\" /etc/sysconfig/kernel; then
      sed -i 's/NO_KMS_IN_INITRD.*/NO_KMS_IN_INITRD="yes"/g' /etc/sysconfig/kernel
      mkinitrd
      fi

      3) Reboot your machine.

      4) Use the nvidia installer for 260.19.36.

      sh NVIDIA-Linux-x86_64-260.19.36.run -q

      IMPORTANT: You need to recompile and install the nvidia kernel module
      after each kernel update.

      sh NVIDIA-Linux-x86_64-260.19.36.run -K

      Michael

      PS: instructions are for SUSE, but I guess they should work everywhere else.

      Team Linux Users Everywhere

      microchip
      microchip
      Joined: 10 Jun 06
      Posts: 50
      Credit: 218927781
      RAC: 193979

      Thanks for the info. Yes, I

      15 Apr 2011 17:39:57 UTC
      Message 105096
      (moderation:
        )

        Thanks for the info. Yes, I know how to manually install the NV driver, but thanks anyways... So I don't need anything else for CUDA on linux? I saw on the NV site a CUDA kit but it's an SDK, not sure if I need that

        Michael Karlinsky
        Michael Karlinsky
        Joined: 22 Jan 05
        Posts: 888
        Credit: 23502182
        RAC: 0

        RE: I saw on the NV site a

        15 Apr 2011 18:01:16 UTC
        Message 105097 in response to message 105096
        (moderation:
          )

          Quote:
          I saw on the NV site a CUDA kit but it's an SDK, not sure if I need that

          This is for development only. All BOINC-project (should) provide the rest (there are some libs, which are necessary).

          Michael

          Team Linux Users Everywhere

          Jeroen
          Jeroen
          Joined: 25 Nov 05
          Posts: 379
          Credit: 740030628
          RAC: 0

          Depending on which GPU you

          15 Apr 2011 22:51:01 UTC
          Message 105098
          (moderation:
            )

            Depending on which GPU you add to your system, you may be able to run multiple work units at once per GPU using an app_info.xml file. Specifically the 4xx and 5xx Fermi GPUs can run two or more work units at a time optimally provided there is enough GPU memory. Each work units requires around 300-350 MB of GPU memory.

            microchip
            microchip
            Joined: 10 Jun 06
            Posts: 50
            Credit: 218927781
            RAC: 193979

            RE: Depending on which GPU

            2 Sep 2011 12:23:43 UTC
            Message 105099 in response to message 105098
            (moderation:
              )

              Quote:
              Depending on which GPU you add to your system, you may be able to run multiple work units at once per GPU using an app_info.xml file. Specifically the 4xx and 5xx Fermi GPUs can run two or more work units at a time optimally provided there is enough GPU memory. Each work units requires around 300-350 MB of GPU memory.

              Can you be more specific about the app_info.xml file? Where do I find it and what do I have to put in it?

              Richard Haselgrove
              Richard Haselgrove
              Joined: 10 Dec 05
              Posts: 2143
              Credit: 2992929455
              RAC: 708360

              I'd recommend that you start

              2 Sep 2011 13:33:13 UTC
              Message 105100 in response to message 105099
              (moderation:
                )

                I'd recommend that you start with the driver stuff on it's own - you don't need an app_info.xml file for that.

                Get it running, check some tasks through and make sure that they validate. The time for moving up to multiple copies is after you've got the plain vanilla version running OK.

                microchip
                microchip
                Joined: 10 Jun 06
                Posts: 50
                Credit: 218927781
                RAC: 193979

                Guys, a final

                8 Sep 2011 14:14:26 UTC
                Message 105101
                (moderation:
                  )

                  Guys, a final question...

                  Do I need a specific driver version or can I grab the 280.13 one?

                  I'm asking because on http://developer.nvidia.com/cuda-toolkit-40 I see there's an older driver version (270.41.19) while the normal driver download page of nVIDIA offers version 280.13... so which one should I get ?

                  Thanks

                  Richard Haselgrove
                  Richard Haselgrove
                  Joined: 10 Dec 05
                  Posts: 2143
                  Credit: 2992929455
                  RAC: 708360

                  You certainly don't need a

                  8 Sep 2011 14:32:30 UTC
                  Message 105102 in response to message 105101
                  (moderation:
                    )

                    You certainly don't need a developer toolkit. Just the ordinary 280.13 should do it, though we have been having some problems with downclocking with the newer *windows* drivers. If you seem to have any problems like that, look for a menu item for "Beta and Archived Drivers" on the general - not developer - download page, and you should be able to find what you need there.

                    microchip
                    microchip
                    Joined: 10 Jun 06
                    Posts: 50
                    Credit: 218927781
                    RAC: 193979

                    OK... thanks Richard for your

                    8 Sep 2011 15:35:41 UTC
                    Message 105103
                    (moderation:
                      )

                      OK... thanks Richard for your help :)

                      Comment viewing options

                      Select your preferred way to display the comments and click "Save settings" to activate your changes.

                      Forums › Cruncher's Corner