CODEWITHSUNDEEP
cudaubuntu-12.04docker
pradeep
pradeep

Reputation: 747

share CUDA library with docker container

I am using docker to launch isolated apps. My app will use CUDA library for GPGPU computing. I am thinking of building an image with cuda installed on it but that would make it too huge. I am not sure either if CUDA display driver installation would go throuh successfully on docker image.

Better way would be to share CUDA libraries on the machine with the container, is there a way to do that ?

Thank you, pradeep.

Upvotes: 0

Views: 2287

Answers (2)

SvenDowideit
SvenDowideit

Reputation: 5208

I would start with just using 

docker run --privileged ubuntu -ti --rm bash

the --privileged is a big hammer, but gives you a container that has full power :)

Upvotes: 3

Andy
Andy

Reputation: 38237

If you can mount your driver as a volume on your host, then you might be able to talk with it directly from within the container by mounting it with docker run -v hostdir:containerdir:rw (see run docs )

If you don't have the device mounted outside the container, then I suspect that talking directly to your graphics card, even through shared libraries, will not work using a standard Docker configuration because the default LXC configuration in docker excludes sys_rawio for both security and portability reasons.

See the lxc_template.go file for the list of capabilities dropped by default.

Upvotes: 1

Related Questions