Add GPU tests to run on Intel GPUs

[motjuste]

The integration-gpu tests assume that the available GPUs are from NVIDIA.

data-science-stack/tests/integration/test_dss.py

Line 70 in 2a10edc

) # This checks for the presence of NVIDIA in the log, assuming only NVIDIA GPUs are used.

Enabling testing on NVIDIA GPUs can be done by adding another tox environment like the one done for integration-gpu, but with a different notebook image:

data-science-stack/tox.ini

Line 108 in 2a10edc

pytest {[vars]tst_path}/integration/ -m gpu -v --tb native -s --notebook-image={env:NOTEBOOK_IMAGE:kubeflownotebookswg/jupyter-pytorch-cuda-full:v1.8.0} {posargs}

Furthermore, tests will have to be added to check using Intel GPUs.

[misohu]

Hi @motjuste sounds like a good idea but the problem is we dont have hardware to develop the tests with Intel. I know we can use testflinger to run a machine in the lab but right now seems like a too much effort.