Enable GPU Legacy Drivers

Feature Requests
,
1

Hello,

Below I write up my suggestion to enable legacy drivers support for legacy GPUs. I list what benefits, advantages and disadvantages I can foresee and present a User Story of a potential user, what error they may encounter and how they would resolve it with the requested feature.

Problem/Justification
Many users, including myself, are using older but still fully functional NVIDIA GPUs that are no longer supported by the default drivers in TrueNAS. For example, my Quadro K4000 requires legacy NVIDIA drivers, specifically version 470. Currently, TrueNAS does not provide an option to select or change the NVIDIA driver being used, which results in compatibility issues. This leads to frequent errors, such as k3s pods crashing and restarting constantly, which hinders the system’s usability. An option to select legacy drivers would allow users to continue leveraging their older hardware without sacrificing system stability.

Impact
This feature would benefit a number of TrueNAS users who are running older GPUs that remain capable of handling their workloads. By allowing users to select or change the NVIDIA driver, TrueNAS could accommodate a broader range of hardware, extending the useful lifespan of legacy GPUs. This would improve system reliability for those affected and reduce the need for hardware upgrades solely due to software limitations. The downside could be added complexity in driver management, but this could be mitigated by making the feature optional or advanced-user focused.

  • Benefits:

    • Increased hardware compatibility and utilization, allowing users with older NVIDIA GPUs to fully leverage their existing hardware.
    • Improved system stability and performance by resolving k3s pod crashes and errors associated with driver incompatibility.
    • Reduced e-waste by extending the lifespan of functional hardware.
    • Enhanced user experience by eliminating persistent error messages and resource consumption caused by pod instability.

  • Advantages:

    • Provides a flexible solution for users with diverse hardware configurations.
    • Empowers users to optimize their systems for specific workloads.

  • Disadvantages:

    • Increased complexity in driver management.
    • Potential for user error if incorrect drivers are selected. This can be mitigated through clear documentation and warnings.

User Story
Sarah, an advanced TrueNAS user, relies on her TrueNAS system for storing and processing large design files. Her TrueNAS setup includes a Quadro K4000, a still-capable GPU that is not supported with the current/default TrueNAS NVIDIA driver configuration.

She navigates to the TrueNAS web UI and locates a new “GPU Driver Management” section under “System → Advanced Settings”. Here, she discovers a dropdown menu listing available NVIDIA drivers, including legacy versions. After selecting the recommended driver (version 470) for her Quadro K4000 and clicking “Apply,” TrueNAS automatically installs the selected driver.

Sarah can get back to running her designer-specific workloads without interruptions. She celebrates with a well-deserved cup of coffee, happy that she didn’t have to spend money on a new GPU.

==========Edit==========
I’ve gathered a list of similar issues (almost entirely from the old forums) which would be greatly helped if the end users were able to somehow select which driver is best for their hardware.

Thank you in advance, team! Love your work!

REFERENCES:

https://ixsystems.atlassian.net/browse/NAS-125436
https://ixsystems.atlassian.net/browse/NAS-122988

1 Like
2

Voted, I also think that ultimately this should be for Nvidia/AMD and Intel GPU’s. There are lots of examples of older hardware that is still more than capable of completing required workloads.

1 Like
3

I agree with you completely! However, for the sake of the feature and the team that would potentially develop it, I would assume it is more digestible if it is delimited to one brand of GPUs. At least to begin with :sweat_smile:

4

Hopefully with the new installable driver model they developed this will be possible

1 Like
5

Just FYI @scyto and @panzerscope , I’ve updated my post with a bunch of links to similar issues people has posted about, spanning years back. All but one of them are from the old forums.

This is all to say that this is not a new issue. Many would be helped immediately and hopefully also many newcomers in the future.

2 Likes
6

I also have this problem, I have an Nvidia GPU that worked fine under jellyfin. I would not want to have to replace a perfectly functional GPU just because of driver versioning.

1 Like
7

My thoughts exactly! Thank you for voting the issue!

It’s great that slowly but surely people are voting, so we can show it is an issue worth addressing!