search

NVIDIA / gpu-rest-engine

A REST API for Caffe using Docker and Go
BSD 3-Clause "New" or "Revised" License
421 stars 94 forks source link

GRE and windows support #29

Closed rperdon closed 6 years ago

rperdon commented 6 years ago

I'm working on doing a demo of our carving tool, GRE and model support. My normal setup would be to have a vm for linux with the model serve answering the carving tool (windows app). I wonder if there would be a way to get it running with windows docker.

So far I'm stuck with the problem of the nvidia-container-runtime bug where it indicates unknown runtime specified nvidia. On the linux side, we added a line to the daemon.json pointing to the nvidia-container-runtime, but I don't think such a container exists on the windows side. It would save a linux install and simplify the deployment of our carving/classification tool if I can have an all windows install.

flx42 commented 6 years ago

Maybe with an Hyper-V VM with DDA (direct-device assignment), you would be running Linux containers on Windows. But I have no experience with this and I'm not even sure that's possible (and it requires Windows Server). On Windows, you would have to get bare metal.

rperdon commented 6 years ago

We have windows 2016 licenses (not sure if my surface book 2 will work well with it) I'd be willing to be a guinea pig to develop this.

flx42 commented 6 years ago

You can look at the many existing issues on the nvidia-docker repository: https://github.com/NVIDIA/nvidia-docker/issues/429 https://github.com/NVIDIA/nvidia-docker/issues/197

I gave you everything I know so far ;) You will have to look at the DDA documentation and assign the discrete GPU to the Hyper-V Linux VM.