nvidia-uvm not loaded (Nvidia 375.39)

I’ve been trying to use Docker with Nvidia and one of the images I’ve used complained about nvidia_uvm. Quick glance revealed it is indeed not loaded:

$ lsmod | grep -i nvidia
nvidia_drm 49152 2
nvidia_modeset 790528 3 nvidia_drm
nvidia 12136448 41 nvidia_modeset
drm_kms_helper 151552 1 nvidia_drm
drm 352256 5 nvidia_drm,drm_kms_helper

Any ideas why nvidia_uvm is not there?
I can get it back with

$ nvidia-modprobe -u -c=0

But I have to do after each reboot and I’d rather find out what’s wrong :) .
I’m on Centos 7 , drivers installed from nvidia packge (not pulled from elrepo).

That stupid suid-root helper ‘nvidia-modprobe’ gets called automatically when ‘nvidia_uvm’ is needed. This kernel module is not loaded at boot time.

And using nvidia’s crude installer when elrepo has properly packaged rpms is also a very bad idea.

so in short - this is normal and I shouldn’t worry, right? Even though in my case nvidia-uvm has not been loaded when needed.

As for elrepo - if it had provided kernel updates along with nvidia drivers for those I’d happily use it. But as we know it is not going to happen (and I can’t use stock and damn old default kernel - it so antiquated it does not pick my Wacom tablet). So I’m stuck with Nvidia installer - and apart from the patch that is needed for kernel 4.10 I haven’t encountered any serious problems on my Centos workstation, neither on my older Fedora install.

What makes you think it was needed but not loaded?