So, I will make this short and to the point. I installed an RTX 5090 without knowing it was not well-supported in UNRAID.

The TL;DR of it all is that I had Claude get it working for me and here are the steps that I had Claude outline.

If there is anything that can be added or adjusted to make this work better, let me know, and I can update the original part to make it easier for others. If this just doesn't work for anyone, let me know that, and I can just dump it all out of here.

Here it is:

RTX 5090 on Unraid - Installing Open Kernel Modules

The Problem

The RTX 5090 (GB202/Blackwell) requires NVIDIA's open kernel modules. The standard Nvidia Driver plugin installs proprietary modules by default, which results in:

- nvidia-smi returning "No devices were found"

- dmesg showing: NVRM: installed in this system requires use of the NVIDIA open kernel modules.

The GPU appears in lspci and the driver loads, but the proprietary modules refuse to initialize the card.

The Fix

1. SSH into your Unraid server

2. Check that the GPU is detected by PCIe:

lspci | grep -i nvidia

You should see the RTX 5090 listed.

3. Confirm the error in dmesg:

dmesg | grep -i nvidia

Look for: NVRM: installed in this system requires use of the NVIDIA open kernel modules.

4. Remove the proprietary driver package:

removepkg $(ls /var/log/packages/ | grep '^nvidia-[0-9]')

5. Download the open source driver package for your kernel:

KERNEL=$(uname -r)

wget -O /tmp/nvos.txz "https://github.com/unraid/unraid-nvidia-driver/releases/download/${KERNEL}/nvos-590.48.01-${KERNEL}-1.txz"

Note: Check https://github.com/unraid/unraid-nvidia-driver/releases for the latest driver version available for your kernel.

6. Install the open source package:

installpkg /tmp/nvos.txz

7. Unload the old modules and load the new ones:

rmmod nvidia_uvm nvidia_drm nvidia_modeset nvidia 2>/dev/null

modprobe nvidia

modprobe nvidia_uvm

8. Verify it works:

nvidia-smi

You should see:

NVIDIA GeForce RTX 5090 | 32607MiB | Driver 590.48.01 | CUDA 13.1

9. Make it persistent across reboots - update the plugin settings:

sed -i 's/driver_version=.*/driver_version=latest_nos/' /boot/config/plugins/nvidia-driver/settings.cfg

Hopefully, this helps others that have a 5090 they couldn't use or intended on getting one until they realized it's not well-supported without some work.

I've spent so much time and money trying to get this thing set up for AI experimentation because I didn't know about certain hardware requirements and interconnections.

If anyone is interested in more about what my setup is, let me know.

Here is a little info on it:

M/B: ASRockRack B650D4U-2L2T/BCM Version 4.01 s/n M80-J5S00800165

BIOS: American Megatrends International, LLC. Version 20.07 Dated 10/17/2024

CPU: AMD Ryzen 9 7950X3D 16-Core @ 4200 MHz

HVM: Enabled

IOMMU: Enabled

Cache: L1 - Cache: 1 MiB, L2 - Cache: 16 MiB, L3 - Cache: 128 MiB

Memory: 128 GiB DDR5 (max. installable capacity 128 GiB)

(Special thank you to JorgeB for suggesting I put this in here.)

Edited February 16Feb 16 by teedge77

You should only need to use the Nvidia Drivers plugin and select the open driver.

22 minutes ago, Kilrah said:

You should only need to use the Nvidia Drivers plugin and select the open driver.

It's entirely possible I just messed it up and then fixed it.

I originally tried the proprietary drivers, and it wouldn't see my 5090 at all. So, I read up on it some and saw I needed the open source drivers. When I tried those, it still didn't see the 5090.

I do remember Claude saying it was still referencing the proprietary drivers or something somewhere. Maybe it didn't remove the proprietary drivers completely when i changed to the open source drivers, or I needed to reboot first. I'm not sure.

I tried a few things and gave up.