K8s's pod fail to find gpu

general condition, we were able to executed nvidia-smi in the pod.

but I attached --cpu-manager-policy=static of kubelet cmd parameter ,after executed nvidia-smi in the pod that show error message failed to initialize NVML: unknow error

we need this parameter. how can we solve it?

system info below.
kubernetes 1.21
nvidia driver 495.46 RTX3090Ti
nvidia-device-plugin 1.10