As it is possible ( intel/intel-device-plugins-for-kubernetes#1686 ) and for me, my only amd64 Node is the one with GPU and has a Taint so only GPU tasks are running on there, it would be quite beneficial to set a toleration to the operator deployment