Enable GPU on Container

@robertgzr
Ok, which image should I flash? (Intel NUC, Up board, MS Surface, etc)

I will sign up to balenaCloud and create a project. Please send me an email to s.carlyon@viso.ai and will send the link.

Cheers,
Shane.

There should be a generic-x86_64 device type that you can use

Cool.

Please give me your email and let me add you as a collaborator to my application.

Device should be ready soon.

Cheers,
Shane.

you shouldn’t have to add me as a collaborator right now. granting support access should suffice.

image

@robertgzr
Ok, I have just enabled support access.

Anything else needed? :slight_smile:

Cheers,
Shane.

Hi there, could you please share the UUID of your device with us here?
Thanks!

@georgiats

8c645a90dcb2e128bb821b10f8ae2914

Thanks!

@georgiats @robertgzr
Were you able to SSH into our device?

This is being investigated by Robert and we will update you when he has some progress.

Hey, guys.
Any update here?

Hi,

We don’t have anything to share yet, we will update you as soon as possible.

hey sorry this took a bit. could you put the device online for me? I can take a look now

@robertgzr
Yes, it’s online now.

alright, so assuming this device has a gpu and ubuntu includes the drivers you need you should be able to build something along the lines of https://github.com/balena-io-playground/jetson-base-images/tree/master/jetson-nano-cuda for your application. Basically you would need to package everything (drivers, system libs, etc.) needed to use the gpu in your container image.

then you can run the container similar to this: https://github.com/balenalabs-incubator/opendatacam (I think privileged is the main takeaway)


we are working on a simpler/nicer solution to this that would move a lot of those requirements (drivers and system libs) into host os space - and make us deal with it ^^
we will definitely make some noise in this thread when that work is ready