Nvidia-smi gives 2 GPUs on 0 GPU machine // Mismatch TF version

Hi there

I am observing a strange behavior when starting a machine with zero GPUs.

base image: ARG RENKU_BASE_IMAGE=renku/renkulab-cuda-tf:11.2-tf-2.7-0.11.0
TF version 2.7.0

  1. If I connect to the machine and I run “nvidia-smi”, I see two GPUs! Instead of zero.
  2. If I run import tensorflow as tf and tf.version in a python session I get 2.9.0 instead of 2.7.0 which the Dockerimage should be built on.

Could you explain what’s going on here? Thx.

See screencast attached here: 2022-06-26_19-05-38.mp4