Linux 64 bit - NVIDIA GPU driver version 304.60 - Errror in cuda application

Forums › Problems and Bug Reports

Bent Vangli
Bent Vangli
Joined: 6 Apr 11
Posts: 23
Credit: 725742660
RAC: 0
19 Oct 2012 10:12:43 UTC
Topic 196578
(moderation:
    )

    All my CUDA applications error out for this driver on a fresh installed Ubuntu 12.04.1 LTS 64 bit version with NVIDIA NVS310 card.

    It seems to be a linking problem to dynamically library according to the error message I find:

    Quote:
    sched_request_einstein.phys.uwm.edu.xml:../../projects/einstein.phys.uwm.edu/einsteinbinary_BRP4_1.28_i686-pc-linux-gnu__BRP4cuda32nv270: error while loading shared libraries: libcuda.so.1: wrong ELF class: ELFCLASS64

    It appears that the program try to load the 64 bit version located in /usr/var/lib, while the correct 32 bit library is located in /usr/var/lib32.

    Is this hardcoded in the program or does it exist a workaround to overcome this problem?

    With very best regards

    Bent Vangli, Oslo, Norway

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

    Linux 64 bit - NVIDIA GPU driver version 304.60 - Errror in cuda

    19 Oct 2012 11:21:14 UTC
    Message 112485
    (moderation:
      )

      Hi,

      make sure you install 32bit drivers *and* 32bit compatibility libs.
      The manual driver install script explicitely asks, if you want to install 32bit too.

      KR
      Michael

      Team Linux Users Everywhere

      Jord
      Joined: 26 Jan 05
      Posts: 2952
      Credit: 5893653
      RAC: 0

      Install the 32bit

      19 Oct 2012 11:25:35 UTC
      Message 112486
      (moderation:
        )

        Install the 32bit compatibility libraries like so: http://boinc.berkeley.edu/wiki/Installing_BOINC#Ubuntu

        Bent Vangli
        Bent Vangli
        Joined: 6 Apr 11
        Posts: 23
        Credit: 725742660
        RAC: 0

        RE: I did find a workaround

        19 Oct 2012 11:28:00 UTC
        Message 112487
        (moderation:
          )

          Quote:

          I did find a workaround to get the get rid of this error:

          Enter the project directory /var/lib/boinc-client/projects/einstein.phys.uwm.edu as root and make a symbolic link to the correct library:

          Quote:
          ln -s /usr/lib32/libcuda.so.1 libcuda.so.1

          However, note that it may happen that the application still doesn't run because of memory lack on the card. NVS310 has only 512 MByte onboard RAM and my application is awaiting more resources :-(

          With best regards Bent

          Bent Vangli
          Bent Vangli
          Joined: 6 Apr 11
          Posts: 23
          Credit: 725742660
          RAC: 0

          I have upgraded to NVIDIA

          19 Oct 2012 14:21:04 UTC
          Message 112488 in response to message 112487
          (moderation:
            )

            I have upgraded to NVIDIA driver version 310.14 beta, and it also seems to work. However, the Einstein CUDA applications aren't able to run despite more than 450 MByte free memory on the card. It complains of too little memory

            I have now tested with Milkyway CUDA, and they seems to run fine. But before I can be conclusive I have to see that the results validate.

            Best regards Bent

            Bent Vangli
            Bent Vangli
            Joined: 6 Apr 11
            Posts: 23
            Credit: 725742660
            RAC: 0

            A correction to my previous

            19 Oct 2012 20:00:37 UTC
            Message 112489 in response to message 112488
            (moderation:
              )

              A correction to my previous mail.

              It was Milkyway OpenCL_NVIDIA that runs. The results done bu NVIDIA driver version 304.60 didn't validate, however the version 310.14 beta gives valid results.

              The beta driver also run Primegrid CUDA applications, but it is too early to see if they validate.

              Best regards Bent

              Bent Vangli
              Bent Vangli
              Joined: 6 Apr 11
              Posts: 23
              Credit: 725742660
              RAC: 0

              Then it is checked. Both

              20 Oct 2012 15:51:56 UTC
              Message 112490 in response to message 112489
              (moderation:
                )

                Then it is checked. Both Milkyway OpenCL-NVIDIA and Primegrid Cuda applications work for NVS 310 mikro GPU, and the results validate. :-)

                Einstein Cuda applications try to start, but after 2 seconds go into hibernation telling it is waiting for more CPU/GPU memory. I do have 16 GByte CPU memory and more than 450 MByte free GPU memory.

                I wonder what the minimum requirements are for the Einstein Cuda applications?

                Best regards Bent

                PS! This is not important but more for fun. The NVS310 is more a standard video card and not intended for heavy GPU crunching :-).

                Gundolf Jahn
                Gundolf Jahn
                Joined: 1 Mar 05
                Posts: 1079
                Credit: 341280
                RAC: 0

                According to BRP CUDA

                20 Oct 2012 16:09:23 UTC
                Message 112491 in response to message 112490
                (moderation:
                  )

                  According to BRP CUDA requirements ( READ ONLY thread ) it seems sufficient, but your BOINC version shows the "read memory bug" for host 5965511 (it shows 134214143MB), which might convert to negative memory available.

                  Gruß,
                  Gundolf

                  Computer sind nicht alles im Leben. (Kleiner Scherz)

                  joe areeda
                  joe areeda
                  Joined: 13 Dec 10
                  Posts: 285
                  Credit: 320378898
                  RAC: 0

                  RE: According to BRP CUDA

                  22 Oct 2012 3:17:08 UTC
                  Message 112492 in response to message 112491
                  (moderation:
                    )

                    Quote:

                    According to BRP CUDA requirements ( READ ONLY thread ) it seems sufficient, but your BOINC version shows the "read memory bug" for host 5965511 (it shows 134214143MB), which might convert to negative memory available.

                    Gruß,
                    Gundolf


                    I find the reported GPU memory in My Computers list to be always wrong but the application runs OK and validates for example:

                    NVIDIA GeForce GTX 670 (6143MB)

                    [joe@jack ~]$ nvidia-smi
Sun Oct 21 20:12:04 2012       
+------------------------------------------------------+                       
| NVIDIA-SMI 4.304.43   Driver Version: 304.43         |                       
|-------------------------------+----------------------+----------------------+
| GPU  Name                     | Bus-Id        Disp.  | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap| Memory-Usage         | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 670          | 0000:01:00.0     N/A |                  N/A |
| 38%   57C  N/A     N/A /  N/A |  39%  800MB / 2047MB |     N/A      Default |
+-------------------------------+----------------------+----------------------+

                    Another is reported: NVIDIA GeForce GTX 560 (5119MB)

                    joe@george:~$ nvidia-smi
Sun Oct 21 20:13:57 2012       
+------------------------------------------------------+                       
| NVIDIA-SMI 4.304.43   Driver Version: 304.43         |                       
|-------------------------------+----------------------+----------------------+
| GPU  Name                     | Bus-Id        Disp.  | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap| Memory-Usage         | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 560          | 0000:01:00.0     N/A |                  N/A |
| 52%   76C  N/A     N/A /  N/A |  73%  747MB / 1023MB |     N/A      Default |
+-------------------------------+----------------------+----------------------+

                    and my all time favorite: NVIDIA GeForce GTX 560 Ti (-2147483648MB)

                    joe@holly:~$ nvidia-smi
Sun Oct 21 20:15:06 2012       
+------------------------------------------------------+                       
| NVIDIA-SMI 3.295.40   Driver Version: 295.40         |                       
|-------------------------------+----------------------+----------------------+
| Nb.  Name                     | Bus Id        Disp.  | Volatile ECC SB / DB |
| Fan   Temp   Power Usage /Cap | Memory Usage         | GPU Util. Compute M. |
|===============================+======================+======================|
| 0.  GeForce GTX 560 Ti        | 0000:01:00.0  N/A    |       N/A        N/A |
|  50%   81 C  N/A   N/A /  N/A |  46%  949MB / 2047MB |  N/A      Default    |
|-------------------------------+----------------------+----------------------|

                    Perhaps having a GPU with negative 2 Peta-bytes of memory is why I can't remember shit.

                    Bent Vangli
                    Bent Vangli
                    Joined: 6 Apr 11
                    Posts: 23
                    Credit: 725742660
                    RAC: 0

                    He-he, that must be the case

                    22 Oct 2012 6:43:51 UTC
                    Message 112493 in response to message 112492
                    (moderation:
                      )

                      He-he, that must be the case LOLOL :-)))). And in an application report I even find that the GPU only had 8 DualPrecision FLOPs/sec. Yes 8 FLOPs, not GFLOPs. It was marked as bizarre low. :-)))))

                      Otherwise Einstein runs flawless on my other computers with ordinary good GPU's.

                      Have a wonderful crunching week, Bent

                      Bent Vangli
                      Bent Vangli
                      Joined: 6 Apr 11
                      Posts: 23
                      Credit: 725742660
                      RAC: 0

                      Today I got my fresh new

                      8 Nov 2012 11:08:45 UTC
                      Message 112494 in response to message 112493
                      (moderation:
                        )

                        Today I got my fresh new computer up and run with the AMD FX8350 CPU, 16Gbyte RAM and GTX 680 GPU card with 4 GByte, Ubuntu 12.10 64 bit and BOINC 7.0.27.

                        Then I happily connected to the Einstein project, and what a disappointment. The standard CPU applications runs just fine, but the GPU versions are telling about too little memory, just as I have with my micro GPU (see the thread).

                        I know this may be a issue with the BOINC client, but all the Primegrid GPU applications runs just fine.

                        I really do wanna crunch Einstein GPU applications, but this silly bug do stop me. Is it possible, either by an extra option on the prefernece page or as a hardcoding to make an override of the memory test? Thus it will be possible to run the application despite this BOINC bug, just as the Primegrid applications do pretty well.

                        With very best regards

                        Bent Vangli, Oslo, Norway

                        PS! My older installations runs both CPU and GPU Einstein application.

                        Comment viewing options

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

                        Forums › Problems and Bug Reports