Does your JupyterHub provide access to GPUs? If so, please tell us more about how you provide it (driver support, software images, GPU resource exhaustion, time-slicing, multi-instance groups, cost management, etc.). Please, share a link to public configuration, where possible.