Login Loop problem Ubuntu 16.04 after installing V100 driver

I have googled this problem but got lost in a jumble of technical try-this try-that’s…

What is the real fix?

Setup is vanilla - GV100 board (SXM module, no display output), fresh clean install of Ubuntu 16.04, on-board generic graphics on a desktop motherboard (Gigabyte), Intel Xeon CPU.

The idea is to build up a backup / clone of the Jetson Xavier we have set up to do Machine Vision for parts inspection.

After installing Volta drivers and rebooting, GUI desktop login appears, but after log in, immediately logs off and again displays the login screen.

I tried disabling and removing the nouveau driver, no joy.

What is the real fix?

This may be a red herring but I have to ask: “GV100 board (SXM module […] on a desktop motherboard”. How are you cooling this thing?

Yes it is a heat monster. I have made a “wind funnel” using polycarbonate sheets and a push-pull dual fan system.

It turns out I can log into this machine remotely, and do all the machine vision and tensorflow stuff I need to, so I guess I don’t really need the lightdm GUI to work. But I would really like to fix this bug if possible.

The “real fix” is to prevent the OpenGL libs from being installed during the driver installation phase of CUDA install.

If you use a runfile installer,

https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#runfile-advanced

pass the switch: --no-opengl-libs

No OpenGL Libraries	--no-opengl-libs	Prevents the driver installation from installing NVIDIA's GL libraries. Useful for systems where the display is driven by a non-NVIDIA GPU. In such systems, NVIDIA's GL libraries could prevent X from loading properly.

Note that this won’t necessarily work if you just reinstall CUDA. You may need to reinstall the OS.

The instructions here are generally sound:

https://devtalk.nvidia.com/default/topic/1046157/linux/ubuntu-16-04-gui-login-loop-after-installing-nvidia-driver/post/5308676/#5308676

I point this out because the first group of instructions there should probably be sufficient to undo the “damage” to the display driver stack that was done, without having to reinstall the OS. However if those steps did not restore graphics operation, if it were me, I would just reinstall the OS.

Those instructions also indicate how to work around this problem using just a driver runfile installer (not the same as a CUDA runfile installer) followed by a package manager CUDA install method that skips the driver install.