yank: "Error launching CUDA compiler: 256" on OpenMM context creation

Keep having this error on the cluster running with nodes=2:ppn=4:gpus=4:shared. The exception is raised after the solvent phase has been completed, during resuming of the complex phase. I’m looking into this.

Traceback (most recent call last):
  File "/cbio/jclab/home/andrrizzi/miniconda/bin/yank", line 9, in <module>
    load_entry_point('yank==0.9.0', 'console_scripts', 'yank')()
  File "/cbio/jclab/home/andrrizzi/miniconda/lib/python2.7/site-packages/yank-0.9.0-py2.7-linux-x86_64.egg/yank/cli.py", line 105, in main
    dispatched = getattr(commands, command).dispatch(args)
  File "/cbio/jclab/home/andrrizzi/miniconda/lib/python2.7/site-packages/yank-0.9.0-py2.7-linux-x86_64.egg/yank/commands/script.py", line 34, in dispatch
    yaml_builder.build_experiment()
  File "/cbio/jclab/home/andrrizzi/miniconda/lib/python2.7/site-packages/yank-0.9.0-py2.7-linux-x86_64.egg/yank/yamlbuild.py", line 1203, in build_experiment
    self._run_experiment(combination, output_dir)
  File "/cbio/jclab/home/andrrizzi/miniconda/lib/python2.7/site-packages/yank-0.9.0-py2.7-linux-x86_64.egg/yank/yamlbuild.py", line 1778, in _run_experiment
    yank.run()
  File "/cbio/jclab/home/andrrizzi/miniconda/lib/python2.7/site-packages/yank-0.9.0-py2.7-linux-x86_64.egg/yank/yank.py", line 451, in run
    simulation.run(niterations_to_run=niterations_to_run)
  File "/cbio/jclab/home/andrrizzi/miniconda/lib/python2.7/site-packages/yank-0.9.0-py2.7-linux-x86_64.egg/yank/repex.py", line 827, in run
    self._initialize_resume()
  File "/cbio/jclab/home/andrrizzi/miniconda/lib/python2.7/site-packages/yank-0.9.0-py2.7-linux-x86_64.egg/yank/repex.py", line 1011, in _initialize_resume
    self.platform = self._determine_fastest_platform(representative_system)
  File "/cbio/jclab/home/andrrizzi/miniconda/lib/python2.7/site-packages/yank-0.9.0-py2.7-linux-x86_64.egg/yank/repex.py", line 901, in _determine_fastest_platform
    context = openmm.Context(system, integrator)
  File "/cbio/jclab/home/andrrizzi/miniconda/lib/python2.7/site-packages/simtk/openmm/openmm.py", line 15103, in __init__
    this = _openmm.new_Context(*args)
Exception: Error launching CUDA compiler: 256
<built-in>:0:0: fatal error: when writing output to : Bad file descriptor
compilation terminated.

About this issue

Original URL
State: closed
Created 8 years ago
Comments: 19 (19 by maintainers)

Most upvoted comments

Some additional testing notes:

This might actually be an MPI version problem. We are using mpich2-1.4.1p1 (not sure which package installs it). However, mpich2 got to version 1.5-ish then changed back in November 2012 the versioning to just mpich starting at 3.0. There is now a 3.2 on conda-forge and it appears to not have this problem on the simple test. I will test with YANK itself first to see if that really is the problem.

Lnaden on Feb 23, 2017