<html>
  <head>
    <meta http-equiv="Content-Type" content="text/html; charset=UTF-8">
  </head>
  <body text="#000000" bgcolor="#FFFFFF">
    Hi Takanori and Dario,
    <br>
    <blockquote type="cite"
      cite="mid:725f1e3d-945e-5887-7475-9afd68b7c73c@meduniwien.ac.at">
      <b>Thanks a lot for your hints!</b><br>
    </blockquote>
    <p>As I pointed out, I can, unfortunately, not change the CUDA
      version neither the nvidia driver, which is currently 410.78 (the
      newest one is 410.93). The parameter "-DCUDA_ARCH=61" did not
      remove the error. <br>
    </p>
    <p>Could it be that I made a mistake (maybe something missing in the
      path?) although I do not get any error on compilation....</p>
    <blockquote type="cite"
      cite="mid:725f1e3d-945e-5887-7475-9afd68b7c73c@meduniwien.ac.at">"echo
      $PATH" gives me:
      <br>
/usr/lib64/openmpi3/bin:/home/blaas/bin/:/home/blaas/software/RELIONCUDA/relion-3.0_beta/build/bin:/usr/local/bin:/usr/bin:/usr/local/sbin:/usr/sbin:/usr/local/cuda/bin/
      and:<br>
    </blockquote>
    <blockquote type="cite"
      cite="mid:725f1e3d-945e-5887-7475-9afd68b7c73c@meduniwien.ac.at">"echo
      $LD_LIBRARY_PATH" gives me:
      <br>
/home/blaas/software/usr/include:/home/blaas/software/usr/lib64:/usr/lib64:/home/blaas/software/RELIONCUDA/relion-3.0_beta/build/lib:/usr/lib64/openmpi/lib:/usr/lib64:/home/blaas/software/RELIONCUDA/relion-3.0_beta/ttt/lib:/usr/lib64/openmpi/lib
      <br>
      <br>
      "nvcc --version" gives me: Cuda compilation tools, release 10.0,
      V10.0.130
      <br>
    </blockquote>
    and "cat /usr/local/cuda/version.txt" gives me: CUDA Version
    10.0.130
    <br>
    <blockquote type="cite"
      cite="mid:725f1e3d-945e-5887-7475-9afd68b7c73c@meduniwien.ac.at">
      <br>
      <b>Do you believe that updating the Nvidia-driver and/or
        downgrading the CUDA version might fix the problem? Does anybody
        use CUDA-10 yet?</b><b><br>
      </b><b>
      </b><br>
      bw Dieter
      <br>
      <br>
      <br>
------------------------------------------------------------------------
      <br>
      Dieter Blaas,
      <br>
      Max F. Perutz Laboratories
      <br>
      Medical University of Vienna,
      <br>
      Inst. Med. Biochem., Vienna Biocenter (VBC),
      <br>
      Dr. Bohr Gasse 9/3,
      <br>
      A-1030 Vienna, Austria,
      <br>
      Tel: 0043 1 4277 61630,
      <br>
      Fax: 0043 1 4277 9616,
      <br>
      e-mail: <a class="moz-txt-link-abbreviated" href="mailto:dieter.blaas@meduniwien.ac.at">dieter.blaas@meduniwien.ac.at</a>
      <br>
------------------------------------------------------------------------
      <br>
      <br>
      Am 05.02.2019 um 08:45 schrieb Takanori Nakane:
      <br>
      <blockquote type="cite">Hi,
        <br>
        <br>
        Which version of nvcc did you use for compilation?
        <br>
        Do you have the right version of CUDA runtime in
        LD_LIBRARY_PATH?
        <br>
        <br>
        Best regards,
        <br>
        <br>
        Takanori Nakane
        <br>
        <br>
        On 2019/02/05 5:20, Dieter Blaas wrote:
        <br>
        <blockquote type="cite">Dear all,
          <br>
          <br>
               I did a test install of the latest version of
          Relion-3.0-beta on a relatively potent workstation (CentOS
          Linux release 7.5.1804) and everything runs fine on all 64
          CPUs. However, when running e.g. Class2D using 2 or 4 GPUs
          (set to 5 MPIs/4 Threads and 3 MPIs/2 Threads, respectively) I
          receive the error below immediately upon starting the run. The
          4 GPUs are 1080 Ti (11 GB). To the best of my knowledge none
          of the possibilities below applies. I am not very familiar
          with compilation and cannot exclude that I did something
          wrong. However, there was no error during the compilation with
          cmake3 followed by make and I previously installed this
          version of relion on two other machines without any issue!
          <br>
          <br>
          Thanks for hints, bw Dieter
          <br>
          <br>
          -------------------------------------------
          <br>
          <br>
          ERROR: unknown error in
          /home/blaas/software/RELIONCUDA/relion-3.0_beta/src/ml_optimiser_mpi.cpp
          at line 128 (error-code 30)
          <br>
          in:
/home/blaas/software/RELIONCUDA/relion-3.0_beta/src/acc/cuda/cuda_settings.h,
          line 67
          <br>
          === Backtrace  ===
          <br>
/home/blaas/software/RELIONCUDA/relion-3.0_beta/ttt/bin/relion_refine_mpi(_ZN11RelionErrorC1ERKSsS1_l+0x41)
          [0x448f81]
          <br>
/home/blaas/software/RELIONCUDA/relion-3.0_beta/ttt/bin/relion_refine_mpi()
          [0x45e513]
          <br>
/home/blaas/software/RELIONCUDA/relion-3.0_beta/ttt/bin/relion_refine_mpi(_ZN14MlOptimiserMpi10initialiseEv+0x2284)
          [0x468a54]
          <br>
/home/blaas/software/RELIONCUDA/relion-3.0_beta/ttt/bin/relion_refine_mpi(main+0xb79)
          [0x4367b9]
          <br>
          /usr/lib64/libc.so.6(__libc_start_main+0xf5) [0x7fc01f87a3d5]
          <br>
/home/blaas/software/RELIONCUDA/relion-3.0_beta/ttt/bin/relion_refine_mpi()
          [0x439b0f]
          <br>
          ==================
          <br>
          ERROR:
          <br>
          <br>
          A GPU-function failed to execute.
          <br>
          <br>
            If this occured at the start of a run, you might have GPUs
          which
          <br>
          are incompatible with either the data or your installation of
          relion.
          <br>
          If you
          <br>
          <br>
               -> INSTALLED RELION YOURSELF: if you e.g. specified
          -DCUDA_ARCH=50
          <br>
                  and are trying ot run on a compute 3.5 GPU
          (-DCUDA_ARCH=3.5),
          <br>
                  this may happen.
          <br>
          <br>
               -> HAVE MULTIPLE GPUS OF DIFFERNT VERSIONS: relion
          needs GPUS with
          <br>
                  at least compute 3.5. You may be trying to use a GPU
          older than
          <br>
                  this. If you have multiple generations, try specifying
          --gpu <X>
          <br>
                  with X=0. Then try X=1 in a new run, and so on. The
          numbering of
          <br>
                  GPUs may not be obvious from the driver or intuition.
          For a list
          <br>
                  of GPU compute generations, see
          <br>
          <br>
          en.wikipedia.org/wiki/CUDA#Version_features_and_specifications
          <br>
          <br>
               -> ARE USING DOUBLE-PRECISION GPU CODE: relion was
          been written so
          <br>
                  as to not require this, and may thus have unforeseen
          requirements
          <br>
                  when run in this mode. If you think it is nonetheless
          necessary,
          <br>
                  please consult the developers with this error.
          <br>
          <br>
          <br>
          _______________________________________________
          <br>
          3dem mailing list
          <br>
          <a class="moz-txt-link-abbreviated" href="mailto:3dem@ncmir.ucsd.edu">3dem@ncmir.ucsd.edu</a>
          <br>
          <a class="moz-txt-link-freetext" href="https://mail.ncmir.ucsd.edu/mailman/listinfo/3dem">https://mail.ncmir.ucsd.edu/mailman/listinfo/3dem</a>
          <br>
        </blockquote>
        <br>
        _______________________________________________
        <br>
        3dem mailing list
        <br>
        <a class="moz-txt-link-abbreviated" href="mailto:3dem@ncmir.ucsd.edu">3dem@ncmir.ucsd.edu</a>
        <br>
        <a class="moz-txt-link-freetext" href="https://mail.ncmir.ucsd.edu/mailman/listinfo/3dem">https://mail.ncmir.ucsd.edu/mailman/listinfo/3dem</a>
        <br>
      </blockquote>
      _______________________________________________
      <br>
      3dem mailing list
      <br>
      <a class="moz-txt-link-abbreviated" href="mailto:3dem@ncmir.ucsd.edu">3dem@ncmir.ucsd.edu</a>
      <br>
      <a class="moz-txt-link-freetext" href="https://mail.ncmir.ucsd.edu/mailman/listinfo/3dem">https://mail.ncmir.ucsd.edu/mailman/listinfo/3dem</a>
      <br>
    </blockquote>
  </body>
</html>