NVIDIA Kernel Module Change in TrueNAS 25.10 - What This Means for You

Announcements
,
1

With the upcoming release of the TrueNAS 25.10 “Goldeye” beta, there have been some changes made to the NVIDIA drivers and supporting modules that are compiled and shipped with TrueNAS.

TrueNAS 25.10 now uses the NVIDIA open GPU kernel modules with the 570.172.08 driver. This enables TrueNAS to make use of NVIDIA Blackwell GPUs - the RTX 50-series and RTX PRO Blackwell cards - which many users have requested support for. [1] [2] [3]

Unfortunately, what NVIDIA giveth, NVIDIA taketh away.

The NVIDIA 50-series Blackwell cards require the use of the new open GPU kernel module, but several of NVIDIA’s older generations of GPUs - including the Maxwell, Pascal, and Volta generations - lack the GPU System Processor (GSP) module on their silicon in order to leverage the open kernel module, and thus will no longer function. This includes the GTX 700-series, 900-GTX series, GTX 10-series, the Quadro M-series and P-series, and Tesla M-series and P-series cards.

A full list of compatible GPUs is available on NVIDIA’s GitHub repository to determine if your card is supported with the new open kernel module.

GPUs that are not compatible with the NVIDIA open kernel module can be isolated and passed through to a VM using TrueNAS Virtualization, and the compatible proprietary/closed-source drivers loaded there. Systems running TrueNAS Apps with GPU acceleration may fail to deploy if a GPU is no longer supported in TrueNAS 25.10. Please ensure that you account for this during upgrade processes - we recommend either replacing the GPU with a supported model or removing the GPU acceleration functionality before upgrading to 25.10.

As an alternative, for transcoding focused workloads, TrueNAS 25.10 includes support for the Intel ARC A-series “Alchemist” GPUs through the i915 open source driver. Support for ARC B-series “Battlemage” GPUs is still pending confirmation of kernel compatibility and functionality with the xe driver required for these GPUs.

3 Likes
2

Bugger - thats a perfectly good P2000 in landfill (probably)

2 Likes
3

It can still be isolated to a VM and used there; that’s my plan for at least one of the Pascal cards I’ve got.

4

How do you remove the Nvidia drivers from TrueNAS?

Cos I just tried and it killed every single app - none of which will run until I tick the install nvidia drivers checkbox again

I have an A310 on the way - but I sort of assume that removing the Nvidia drivers is a ghood first step. Only one of my apps actually uses the P2000 (Plex), in theory nothing else does

After re-checking the Nvidia checkbox - they all work again

5

Not sure on that part, exactly. The checkbox removes the drivers, but will also have to stop and start several dependent Docker containers as well - IIRC there are arbitration/control containers and components of the Docker engine that are removed and re-added, so that job and the subsequent start of the Apps would have to happen again.

You’d want to unconfigure the GPU in Plex certainly before pulling the drivers out.

6

I removed the GPU from Plex - that didn’t cause an issue.

What did cause an issue was unchecking the “load nvidia drivers” after which the apps all vanished for a minute and then came back crashed. They all failed when I tried to start them

I may try again tomorrow - but this time with a NAS reboot before restartingthe apps

7

Well a reboot certainly helped.
Of the 11 apps, jellyfin, nginx-proxy-manager & tdarr all failed to start.

Now jellyfin & tdarr MIGHT have had the GPU assigned (I didn’t check as its not used in Tdarr & I forgot jellyfin) - but not nginx. However - editing the app configuration and just clicking update, without changing anything solved the issue - all apps running again.

So after removing the nvidia drivers - reboot and any recalcitrant apps - just edit and update

or thats what my testing shows

8

Even though it would increase the maintenance burden, why not offer a choice of versions? Provide two: one being the last driver release that does not support the open kernel module, and the other being the latest release that does support it.

2 Likes
9

Here’s how to build your custom NVIDIA driver systemd-sysext extensions.

PS: You still have to ensure kernel compatibility.

1 Like
10

Ah, glad I did a search before starting a new thread. Now I know why my Quadro P400 is not working in the Goldeneye Beta. :slight_smile:

Can anyone point me to a similar low-profile card I can slot into my JONSBO N1 that is also supported? Otherwise I guess I can upgrade to a bigger case.

EDIT: It looks like the Nvidia T400 is a good option from the provided list.

1 Like
11

I’m using an nvidia t400 which, afaik is the newer model of the p400 and is supported according to the list of compatible gpus

3 Likes
12

Perfect! That’s the card I’m going with. Thank you!

EDIT: Yeah, the Intel A310 slaps.

13

Interesting, I’m just gonna attempt to install the regular drivers for my 1080ti using standard methods since I have to use install-dev-tools anyways for my Google Coral

14

The T400, as the letter implies, is based on the NVIDIA Turing generation chip, and will work in 25.10. :slight_smile:

Alternatively, you can get an Intel ARC A310, if you have no allegiance to Team Green …

2 Likes
15

Biggest issue for Ollama users is that it doesn’t support Intel GPU’s, so a drawback for anyone using older nvidia cards for AI stuff

16

Pretty sure that there’s a fork of Ollama with Intel support.

Maybe I can bug @kris to talk about his, although I don’t know if he’s still using it? :wink:

17

I don’t want my 1080 card to end up in a land fill. Is there a GPU mounting case that I can purchase to mount that GPU on the wall?

As for a replacement GPU… looks like a 5060 will be a slightly faster replacement.

18

Yeah I know of the fork, helped a guy on Discord about it, see Discord, but he was confused why his CPU was getting hit hard when using the Catalog app, just an FYI on a searchable thing (aka here) in case people are wondering

19

No, I gave up and went with an nvidia card locally. The forked version when I was testing was just too unstable and inconsistent. Granted, that was 6+ months ago, so could be a different situation now.

20

I tried compiling the Google Coral module, but upstream GitHub hasn’t merged fixes for newer kernels, so I gave up (I use a local runner to auto-build each TrueNAS release). Manual patch and compile should still work. Using install-dev-tools is too much hassle after every update, while systemd-sysext just needs linking and merge to work.