[solved] Headless nVidia system?

[ExplodingLemur]

I've got a box in a datacenter with Ubuntu and a GPU in it, and I'd like to run FAH on it without installing a full GUI desktop environment (which the nVidia drivers from Ubuntu repos want to do).
How do I go about this?

[bruce]

The Ubuntu server addition comes without the full desktop GUI. Though I've never tried it myself, I believe you can install the GPU drivers from NVidia without installing the desktop.

[ExplodingLemur]

Ubuntu's nVidia driver package is dependent on the desktop environment packages.

[ExplodingLemur]

Ugh I'm an idiot. I misread the list of dependencies with the Ubuntu package. No, it doesn't pull in a complete desktop environment, just some Gnome libraries. It's installed now. However I'm having issues with the nVidia driver refusing to work with a PCI passthrough card in Proxmox. Working on that now...

[bruce]

I'm not sure about the passthrough card but as a general rule, GPU's the OpenCL environment (which is required by FAH) does not work in a virtual system.

[katakaio]

At the risk of necroing this thread, I found myself in the same boat (Linux box seeking headless NVIDIA install) and spent days figuring it out. I can now share two other solutions to anyone who might be in a similar situation.

The repositories for Debian-based distros like Ubuntu and Linux Mint now include a nvidia-headless-XXX driver (where XXX indicates the version). This will allow you to install a driver that doesn't require X, Wayland, or anything graphical.

This would've been a great solution for me, but the only way that I could control my GPU fan was through the NVIDIA XServer utility, which (you guessed it) requires that the GPU is attached to an X screen. There are many how-tos out there for attaching a fake X screen to your GPU via xorg.conf and I won't repeat them here...but I will say that the driver install that ultimately worked for me was downloading the .run file directly from NVIDIA and installing with the --no-opengl-libs option. A little monkeying with my xorg.conf file and my old 960 is back in action and folding for good

Hopefully one of these solutions will work for folks who are either pursuing a headless install or trying to drive their display with an iGPU and dedicate their GPU to folding. Good luck!

[bruce]

... or trying to drive their display with an iGPU and dedicate their GPU to folding. Good luck!

What do I need to do to enable the iGPU when I have discrete GPU(s)? Do I need to upgrade my BIOS or is there a trick somewhere that I haven't found?

[katakaio]

In my limited experience, the iGPU on most desktop computers is enabled even if a discrete GPU is present with display attached. But if not, you can always peek into the BIOS and look for iGPU settings to fiddle with. On my Intel board, my options are "Auto", "Enabled", and "Disabled" for the iGPU on my Skylake processor. When I decided to use my discrete GPU for folding and my integrated graphics for display, I toggled this from "Auto" to "Enabled" just to play it safe. There was also a setting for init display that I also switched to iGPU from PCIe (i.e. my discrete GPU).

Now if you've been using your discrete GPU for display and you want to switch to using your iGPU, that process will vary by OS. This is one of those situations that Windows seems to handle much better than Linux, in my opinion! A painless task on Windows 7 (it was plug and play for me), but it involved minor witchcraft in Linux Mint.