11

I have a Dockerfile based on nvidia/cuda like so:

FROM nvidia/cuda:11.0-base

...

I want to be able to build this Dockerfile on our CI server that does not have a Nvidia GPU. When I try to do that, I get this error:

------
 > [1/6] FROM docker.io/nvidia/cuda:11.0-base:
------
failed to solve with frontend dockerfile.v0: failed to solve with frontend gateway.v0: rpc error: code = Unknown desc = failed to build LLB: failed to load cache key: docker.io/nvidia/cuda:11.0-base not found

The error says that the image is not found, but I think this is a bit misleading. I've been able to isolate the problem to whether or not a GPU is present.

When building this Dockerfile on a server with a Nvidia GPU, I don't get this error. Is it possible to build a Dockerfile based on an nvidia/cuda image on a server without a GPU? This would save costs on our CI server.

I plan to deploy the resulting docker container on a server that does have a GPU so, in other words, is it possible to defer the presence of a GPU to run time instead of build time?

Mario Ishac
  • 5,060
  • 3
  • 21
  • 52
  • Could you try `docker pull nvidia/cuda:11.0-base` on your ci server? – anlgrses Aug 10 '20 at 06:00
  • It doesn't seem like a GPU issue, could you check if any mirrors are configured on the CI server by doing `docker info`. Its possible different mirrors are configured for GPU servers. – Palash Bansal Aug 16 '20 at 15:31

1 Answers1

1

It sounds like you may need to load the nvidia components possibly including any proprietary blobs and kernel modules. If the modules are not present, this could be why the compile error (missing dependencies).

But from this website https://docs.nvidia.com/datacenter/tesla/tesla-installation-notes/index.html it looks like the drivers are looking for the hardware when they load, which is probably why they are not available when you attempt to compile.

Dharman
  • 30,962
  • 25
  • 85
  • 135
Hmbl Stdnt
  • 11
  • 3