Nvidia and Linux (Yikes)

Update: Screw it, go with AMD. This driver hell is ridiculous.

Yeah, Nvidia is a mess to deal with. That is, regarding Linux video drivers, and getting things working. Graphics on Linux can be nearly as frustrating as audio on Linux (well, not that near but still, frustrating).

For Nvidia video cards, there are several ways this can go wrong, and everything needs to be right to go right. Especially for something like an eGPU. Here are some of the various ways Nvidia and Linux can go wrong. This is about Nvidia, but it applies to other graphics card vendors to some degree. Note: I use Debian where possible.

Bios settings

  • Enough power?
  • Disable sleep modes?
  • Enable Thunderbolt port and / or set security setting on it
  • Disable integrated graphics

Nvidia drivers

  • Linux kernel headers needed
  • GCC needed
  • Drivers and only certain kernels are compatible
  • For the kernel extensions there are the proprietary and open source versions

Cuda drivers

Setting X11 / Xorg

  • PCI slot (decimal to hex)

Disable video drivers

Some handy commands:

  • nvidia-smi - Can Nvidia software talk to the driver
  • uname -r - what Linux version
  • nvtop - GPU utilization realtime graph (akin to htop)
  • lspci | grep "VGA" - what graphics drivers are loaded
  • ``
  • ``

Some handy locations

  • /etc/modprobe.d - where placklist files and dkms.conf is located
  • /usr/share/X11/xorg.conf.d - where various conf files are located, including possibly 10-nvidia.conf and others
  • ``

Some handy

We use cookies only for WordPress functionality, and do not use third-party analytics or tracking.
Accept