installing hoomdgpu on cluster with different versions of cuda

[zahraAkbari8]

Hello
I have installed hoomd gpu with cuda 11.8 on cluster but when running my code i get the following error sometimes:

 gpu = hoomd.device.GPU()
          ^^^^^^^^^^^^^^^^^^
  File "/nfs/scistore18/hannegrp/zakbari/micromamba/envs/hoomd5GPU/lib/python3.11/site-packages/hoomd/device.py", line 356, in __init__
    self._cpp_exec_conf = _hoomd.ExecutionConfiguration(
                          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: CUDA Error: uncorrectable ECC error encountered before /hoomd/ExecutionConfiguration.cc:289

I tried to install hoomd gpu with cuda 12 using https://hoomd-blue.readthedocs.io/en/latest/installation.html but /i get this error:

Could not solve for environment specs

The following packages are incompatible
├─ cuda-version 12.0**  is requested and can be installed;
└─ hoomd 5.0.0 *gpu* is not installable because there are no viable options
   ├─ hoomd 5.0.0 would require
   │  └─ cuda-version >=11,<12.0a0 , which conflicts with any installable versions previously reported;
   └─ hoomd 5.0.0 would require
      └─ __cuda >=12.6 , which is missing on the system.

I also followed the instructions on https://glotzerlab-software.readthedocs.io/en/latest/build.html which causes the error:

# >>>>>>>>>>>>>>>>>>>>>> ERROR REPORT <<<<<<<<<<<<<<<<<<<<<<

    Traceback (most recent call last):
      File "/nfs/scistore18/hannegrp/zakbari/miniforge3/envs/buildforge/lib/python3.12/site-packages/conda/exception_handler.py", line 18, in __call__
        return func(*args, **kwargs)
               ^^^^^^^^^^^^^^^^^^^^^
      File "/nfs/scistore18/hannegrp/zakbari/miniforge3/envs/buildforge/lib/python3.12/site-packages/conda/cli/main.py", line 61, in main_subshell
        exit_code = do_call(args, parser)
                    ^^^^^^^^^^^^^^^^^^^^^
      File "/nfs/scistore18/hannegrp/zakbari/miniforge3/envs/buildforge/lib/python3.12/site-packages/conda/cli/conda_argparse.py", line 181, in do_call
        result = plugin_subcommand.action(getattr(args, "_args", args))
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
      File "/nfs/scistore18/hannegrp/zakbari/miniforge3/envs/buildforge/lib/python3.12/site-packages/conda_build/plugin.py", line 17, in build
        return execute(args)
               ^^^^^^^^^^^^^
      File "/nfs/scistore18/hannegrp/zakbari/miniforge3/envs/buildforge/lib/python3.12/site-packages/conda_build/cli/main_build.py", line 590, in execute
        api.build(
      File "/nfs/scistore18/hannegrp/zakbari/miniforge3/envs/buildforge/lib/python3.12/site-packages/conda_build/api.py", line 250, in build
        return build_tree(
               ^^^^^^^^^^^
      File "/nfs/scistore18/hannegrp/zakbari/miniforge3/envs/buildforge/lib/python3.12/site-packages/conda_build/build.py", line 3638, in build_tree
        packages_from_this = build(
                             ^^^^^^
      File "/nfs/scistore18/hannegrp/zakbari/miniforge3/envs/buildforge/lib/python3.12/site-packages/conda_build/build.py", line 2506, in build
        utils.check_call_env(
      File "/nfs/scistore18/hannegrp/zakbari/miniforge3/envs/buildforge/lib/python3.12/site-packages/conda_build/utils.py", line 405, in check_call_env
        return _func_defaulting_env_to_os_environ("call", *popenargs, **kwargs)
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
      File "/nfs/scistore18/hannegrp/zakbari/miniforge3/envs/buildforge/lib/python3.12/site-packages/conda_build/utils.py", line 381, in _func_defaulting_env_to_os_environ
        raise subprocess.CalledProcessError(proc.returncode, _args)
    subprocess.CalledProcessError: Command '['/bin/bash', '-o', 'errexit', '/tmp/conda-build-zakbari.G6jZ4Shu/hoomd_1734486724742/work/conda_build.sh']' returned non-zero exit status 1.

What do you think is the best way to install hoomd gpu in this case?

Thanks in advance

[joaander]

1. "uncorrectable ECC error" indicates a problem with the hardware itself, not the software configuration.

2. Note the installation error message indicates that CUDA 12.6 is missing. Force it to be available with:

export CONDA_OVERRIDE_CUDA="12.6"
micromamba install "hoomd=5.0.0=*gpu*" "cuda-version=12.6"

3. You shared a meaningless portion of the error output. I can offer no insight.

Next time, try asking 1 question at a time. GitHub Discussions allows threaded discussions.

[zahraAkbari8]

Thanks. But I was surprised that hoomd 5 is not installable with CUDA 12.0 although it is explicitly mentioned in the documentation: https://hoomd-blue.readthedocs.io/en/latest/installation.html

[joaander]

conda-forge used to build with CUDA 12.0, now it builds with 12.6. We do not actively unit test every single block of example code in the documentation and therefore rely on users to report errors.