Hi,We have been trying to experiment with the docker universe, and have found that when a GPU job is requested, the required CUDA device files are not passed through to the container as required [1]
In our case a "--device /dev/nvidiaN:/dev/nvidiaN" per card plus the nvdiactl and uvm ones would be needed.
I think we could use a docker command wrapper to add them as needed [2], but I was wondering if there are plans for a generic support for this kind of features (it needn't be GPU specific, it could be made so that other machine resources are passed through to the docker containers).
Best, Joan[1]: http://stackoverflow.com/questions/25185405/using-gpu-from-a-docker-container [2]: https://www-auth.cs.wisc.edu/lists/htcondor-devel/2015-October/msg00002.shtml
Attachment:
smime.p7s
Description: S/MIME Cryptographic Signature