There has been quite a few requests of running NFD worker pod on the tainted nodes. Usually customers taint specific worker nodes in order to preserve them for GPU workload only, but they still need NFD worker pods to run on those nodes. By default,NFD does not have (yet) support, but in that case we have a workaround.
"In case customer wants to taint the worker node so that only a specific workload with toleration will run on it, but also needs the NFD worker pods to run on the node, then the taint's action should be NoSchedule. In that case NFD worker pod will be able to be scheduled and run on the tainted node"

    1.
    2.
    3.