Does docker need anything special for GPU instances? As far as I know, rancher 1.X simply interacts with the docker agent running on whatever hardware you provide (cloud or otherwise).
Kubernetes version 1.7 (I believe) introduced native GPU support, in beta.
So there are some 'special' things that need to be done...though the gap seems to be getting more narrow.
We have some customers who are using this configuration (including nvidia-docker) with some scheduling limitations.
That being said, we haven't expressly integrated this use case, so anything Kubernetes has done in 1.7 is available, but not completely surfaced/taken advantage of.
Any chance of some documentation/HOWTO on how that was done?
Thanks again!