Something went wrong on our end
Select Git revision
README.md
-
Marty Kandes authored
Including its base and intermediate layers. Again, the MLNX-OFED and OpenMPI layers no longer update all of the software within the container by default near the beginning of the %post section. It was observed with this set of containers in particular that an unintended update of the minor version of the NCCL libraries, which caused a runtime problem when subsequentely used for Horovod + TensorFlow. Apparentely there was a runtime API change in CUDA 11.3.
Marty Kandes authoredIncluding its base and intermediate layers. Again, the MLNX-OFED and OpenMPI layers no longer update all of the software within the container by default near the beginning of the %post section. It was observed with this set of containers in particular that an unintended update of the minor version of the NCCL libraries, which caused a runtime problem when subsequentely used for Horovod + TensorFlow. Apparentely there was a runtime API change in CUDA 11.3.