Running Container Lost GPU Permission

How to reproduce the issue:

1. Create a job on the platform and run nvidia-smi in a infinite loop, it works well
2. Run the sudo systemctl daemon-reload on the VM in the job
3. The job will show 'Failed to initialize NVML: Unknown Error' failure

Related Issues:
https://github.com/NVIDIA/gpu-operator/issues/485
https://github.com/NVIDIA/nvidia-container-toolkit/issues/48