Skip to content

Version 6.2.0

Compare
Choose a tag to compare
@roclark roclark released this 18 May 16:44
· 17 commits to main since this release
3cc4010

Enhancements

  • Add and document common non-zero exit codes (#25)
  • Added parser for mdtest results (#41, #42)
  • Added a tool to ensure all passed hosts are unique to avoid running multiple tests on the same host (#46)
  • Added instructions on running Bobber on non-DGX platforms (#56)
  • Added steps for installing via nvidia-pyindex (#51)
  • Annotated the analysis sub-module (#43)
  • Added baseline for DGX A100 PODs up to 8 nodes (#52)
  • Augmented Docker error handling (#48)
  • Allow pauses between tests to allow filesystem cleanup to occur (#44)
  • Automatically generate a baseline file based on parsed results (#55)
  • Optionally save JSON data from the parser to reduce bloat (#62)
  • Added GitLab Continuous Integration support to run tests against GPUs (#75)

Fixes

  • Removed the NCCL-max default from the parser to apply the system defaults (#64)
  • Fixed the filename for FIO IOPs tests (#72)
  • Fixed an issue creating TFRecords with Imageinary (#76)

Dependency Updates

  • Upgraded Imageinary to 1.1.2 (#59)

Known Issues

  • Recent versions of docker-py throw errors as they rely on a dependency to be installed that isn't listed. To workaround this, either upgrade to version 6.3.0 or newer of Bobber or manually install six.