support nvidia runtime

[kyr7]

Support nvidia runtime

Hi. I want to be able to use nvidia runtime like this:
'docker run --runtime=nvidia nvidia/cuda nvidia-smi'

Would you like to add this feature valuable for machine learning? Or maybe would you hint how I could add it to make a pull request?

[mattnworb]

this should already be possible, check out #1060

[kyr7]

Thank you for pointing this.