In the documentation, we should specify that using GPUs on non CUDA based images (i.e. minimal python and standard data science as of now that we provide, potentially all ISV images?) is unsupported.

If a user were to spawn one of these images with a GPU attached and use e.g. PyTorch, they'd still be able to make use of the hardware, but the same wouldn't hold true for Tensorflow. If we are not explicitly preventing the user from running into this scenario, we should at least specify in the documentation that this is not supported by Red Hat.