How are nvidia tools mounted in containers?

I have been using containers from nvcr.io (like nvidia/cuda:11.0.3-base-ubuntu20.04) and docker hub (like deepops/nccl-tests-tf20.06-ubuntu18.04) and run them with docker or enroot (via slurm). I find that there are some binaries from the host system mounted in these containers like this

/dev/mapper/ubuntu–vg-ubuntu–lv on /usr/bin/nvidia-smi type ext4 (ro,nosuid,nodev,relatime)

When I run these containers with singularity (via slurm) these mounts are missing. Mostly for educational purposes I want to know how these mounts are done and why singularity behaves differently from docker/enroot.

thanks