REMD with GPU version of spdyn

Viewing 7 reply threads
  • Author
    Posts
    • #16454

      geng
      Participant

      Hi all,

      I tried to install the GPU verison of spdyn. However, I failed to pass the test of REMD, while other tests passed. I confused whether the GPU spdyn cannot do REMD or my installation is not correct?

      Best Regards,

      Geng

       

    • #16456

      ckobayashi
      Moderator

      Hi Geng,

       

      REMD in spdyn should work as well as cMD in GPU version.

      Could you share error logs of the test of REMD?

       

      Thank you.

       

      Chigusa

      • #16459

        geng
        Participant

        Sorry for so many messages.

        I tried to use 4/8 parallel cpu cores. One GPU was used.

        /Geng

      • #16466

        geng
        Participant

        We now found that the GPU spdyn failed in REMD because of the cutoff : “Read Ctrl Energy: cutoff is not available with GPU”.

        I do not understant this “cutoff”.

        Best,

        Geng

         

    • #16457

      geng
      Participant

      Hi Chigusa,

      <span style=”text-decoration: underline;”>The final report of test is</span>

      Passed 0 / 10
      Failed 0 / 10
      Aborted 10 / 10

      <span style=”text-decoration: underline;”>I then check the log file in tests-1.6.1/regression_test/test_remd_common/H-REMD_DIHED, and found </span>

      ===================================================================================
      = BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
      = RANK 0 PID 45649 RUNNING AT g0001
      = KILLED BY SIGNAL: 9 (Killed)
      ===================================================================================

      ===================================================================================
      = BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
      = RANK 1 PID 45650 RUNNING AT g0001
      = EXIT STATUS: 1
      ===================================================================================

      ===================================================================================
      = BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
      = RANK 2 PID 45651 RUNNING AT g0001
      = KILLED BY SIGNAL: 9 (Killed)
      =========================================================================

      .

      .

      .

      ===================================================================================
      = BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
      = RANK 14 PID 45663 RUNNING AT g0001
      = KILLED BY SIGNAL: 9 (Killed)
      ===================================================================================

      ===================================================================================
      = BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
      = RANK 15 PID 45664 RUNNING AT g0001
      = EXIT STATUS: 1
      ===================================================================================

      Thank you!

      Best Regards,

      Geng

    • #16458

      geng
      Participant

      I tried again, and nothing was found in log file but

      “/data/apps/intel/2019/compilers_and_libraries_2019.3.199/linux/mpi/intel64/bin/mpirun: line 103: 51908 Segmentation fault mpiexec.hydra “$@” 0<&0″

      was shown in error file.

      Best,

      Geng

    • #16460

      tmori
      Moderator

      Could you try to execute spdyn in the subdirectory “./test_remd_common/H-REMD_DIHED”
      without using “test_remd.py”?

      For example, please type the following commands:
      $ export OMP_NUM_THREADS=1
      $ mpirun -np 8 /PATH_TO_GENESIS/bin/spdyn inp > test.log

      If you could obtain output files, the problem might lie in the python script or command rather than GENESIS.

      Bests,
      Takaharu Mori

      • #16461

        geng
        Participant

        The test.log file is empty, and I found “/data/apps/intel/2019/compilers_and_libraries_2019.3.199/linux/mpi/intel64/bin/mpirun: line 103: 68726 Segmentation fault mpiexec.hydra “$@” 0<&0′ in the slurm-55692.out file.

        Best,

        Geng

      • #16462

        geng
        Participant

        Does it matter that I used the GPU card of V100?

        Best Regards,

        Geng

    • #16463

      tmori
      Moderator

      As shown in the user manual (p. 18 “Run the additional tests”),
      there is an example command for GPU tests:

      $ ./test.py “mpirun -np 8 ~/genesis/genesis-1.6.1/bin/spdyn” gpu

      Did you successfully finish this test?

      Bests,
      Takaharu Mori

      • #16464

        geng
        Participant

        Yes, we have successfully done the test.py.

        I have successfully done the test_rpath.py, test_gamd.py and test_fep.py, but failed for test_remd.py.

        Best,

        Geng

      • #16465

        geng
        Participant

        Does it matter that I used the GPU card of V100, which is not included in the manual (NVIDIA K20, K40, P100, TITAN V, GTX 1080, GTX 1080Ti, RTX 2080, RTX 2080Ti)

        Best Regards,

        Geng

    • #16467

      tmori
      Moderator

      I think in the regression tests for REMD, GPU version of spdyn is not available.
      This is simply because we did not prepare such test options. We are sorry for the confusion.
      Of course, you can perform REMD simulations with GPU spdyn.
      Basically, GPGPU calculation in spdyn requires “electrostatic = PME” (not CUTOFF).

      Best regards,
      Takaharu Mori

    • #16468

      tmori
      Moderator

      For the REMD test using GPU version of spdyn,
      “regression_test/test_remd_spdyn/T-REMD_RESPA” is available.
      Please try this test set.

      Thank you,
      Takaharu Mori

      • #16469

        geng
        Participant

        Thank you for your information.

        I tried T-REMD_RESPA, and it passed.

        Thank you.

        Best,

        Geng

      • #16473

        geng
        Participant

        I can now run remd in gpu version. However, the GPU utilization rate is much low of 25-30%.

        I run simulation with one GPU (V100) and 8 cores of CPU.  I tried with more cpu cores and less cpu cores, and these will lead to ever lower utilization rate of GPU.

        Is it caused by the compilation, e.g. cuda version, intel compiler?

        Do you have any suggestion for this? How do you compile Genesis?

        Thank you very much!

        Best,

        Geng

        • #16474

          jwjung
          Moderator

          Dear Geng,

          In GENESIS, we use GPU for real-space nonbonded interactions (van der Waals and electrostatic) while CPU is used for other parts (bond/angle/dihedral angle calculation and recprocal-space electrostatic). In fact, the performance of GPU is limited by CPU calculation of such parts and I guess it makes the low utilization of GPU.

Viewing 7 reply threads

You must be logged in to reply to this topic.