[BUG] Intel DRI3 Truenas Scale - No GPU Passthrough

[Glujaz]

Is there an existing issue for this?

• I have searched the existing issues

Current Behavior

When moving the 3D part, the GPU is not used. The rendering is done by the CPU

Expected Behavior

The container is installed on a Truenas Scale system, with an intel N150. The GPU is working fine inside truenas, and a jellyfin docker. Doing a "intel_gpu_top" shows activity for jellyfin docker, but not for the Bambulab docker.
The expected behavior is to use the integrated GPU of the CPU to have a smooth experience for basic parts on a slow CPU.

I think it might be related to a not updated GPU driver of some sort.

Steps To Reproduce

Install truenas scale on an N150 system.
Add custom app, select the correct settings, and at the end, no hardware limitation, and enable GPU passthrough.

Passing /dev/dri/ to the docker container, passing the gpu via the environment variables does not help either.

Environment

- OS:Truenas Scale 25.04.2
- How docker service was installed:Custom Apps from Truenas Scale GUI

CPU architecture

x86-64

Docker creation

N/A

Container logs

2025-08-05 19:00:02.899255+00:00MESA: warning: Driver does not support the 0x46d4 PCI ID.
2025-08-05 19:00:02.899315+00:00libEGL warning: egl: failed to create dri2 screen
2025-08-05 19:00:02.905416+00:00MESA: warning: Driver does not support the 0x46d4 PCI ID.
2025-08-05 19:00:02.906827+00:00libEGL warning: egl: failed to create dri2 screen
2025-08-05 19:00:02.931443+00:00MESA: warning: Driver does not support the 0x46d4 PCI ID.
2025-08-05 19:00:02.932427+00:00libEGL warning: egl: failed to create dri2 screen
2025-08-05 19:00:02.947143+00:00MESA: warning: Driver does not support the 0x46d4 PCI ID.
2025-08-05 19:00:02.948207+00:00libEGL warning: egl: failed to create dri2 screen
2025-08-05 19:00:03.004812+00:00MESA: warning: Driver does not support the 0x46d4 PCI ID.
2025-08-05 19:00:03.005938+00:00libEGL warning: egl: failed to create dri2 screen
2025-08-05 19:00:03.032509+00:00MESA: warning: Driver does not support the 0x46d4 PCI ID.
2025-08-05 19:00:03.033568+00:00libEGL warning: egl: failed to create dri2 screen
2025-08-05 19:00:03.048630+00:00MESA: warning: Driver does not support the 0x46d4 PCI ID.
2025-08-05 19:00:03.049530+00:00libEGL warning: egl: failed to create dri2 screen

[github-actions]

Thanks for opening your first issue here! Be sure to follow the relevant issue templates, or risk having this issue marked as invalid.

[thelamer]

DRI3 just works on my AMD card, there is not much else I can do to help you here, if you simply mount with --device /dev/dri:/dev/dri it should just work.

I did buy a crappy $100 n150 (unrelated) I get on Saturday though so I guess it is a race against time for you to figure out what you may potentially have misconfigured and me running this on my local hardware. ( I will be installing Debian Trixie though)

[Glujaz]

In truenas, there is an option to pass the gpu. It is supposed to do as you said. He gpu is mounted inside the docker, I can see and confirm it.
However, the app is really not using it.
I also tried to manually pass through the storage /dev/dri, it is again visible, but nothing is happening as well.

This is why I think the container needs a driver update of some sort.
I have to admit though I am not very familiar with docker.

Could it also be a right thing ? I see that /dev/dri directory and content is read by everyone though, so the docker container should be able to work with it !

[j0nnymoe]

Try deploying the container via docker compose and not using the truenas gui to see if the same thing happens.

[thelamer]

I am not calling this a speed demon but it works.

bambu-n150-gpu.mp4

This is all I ran to test:

docker run --rm -it -p 3001:3001 --shm-size=1gb --device /dev/dri:/dev/dri linuxserver/bambustudio bash

And without it:

bambu-n150-nogpu.mp4

docker run --rm -it -p 3001:3001 --shm-size=1gb linuxserver/bambustudio bash

You go from like 3-5 fps to 25 fps, it is not buttery smooth but it is usable.

I did a bunch of debugging on this n150 today, part of the problem with the platform is all these mini pcs are single channel like 2400mhz ram and we use a virtual framebuffer in docker, so to get the pixmaps from the gpu to the virtual framebuffer to be rendered you need the memory pipe. When I was running it with the gpu the system load is massively less and you get better performance, but it cannot push it to the limit because the memory bandwidth is just not there.

Just letting you know this so you can temper your expectations when you get get it mounted up right.

[Glujaz]

Thanks a lot for the testing.

I am on the move right now and will not be able to Check this again until tuesday.

But this is what i am actually expecting for results, something usable 😅

[Glujaz]

So, I tried to use a YAML config, and also to run the command you showed in truenas.

It seems to work partially. Looking at the "intel_gpu_top" command, I now see that there is Xvnc. But this is just the internal VNC tool I guess, not the GUI itself, so not Bambulab.

Just out of curiosity, how is your motherboard managing a preview after slicing a plate ?
My experience is that it is very slow once it is previewing the slicing result.

[Glujaz]

Odly enough, I get these errors in the log all the time :

2025-08-19 20:27:44.936195+00:00(bambu-studio:326): GLib-GObject-WARNING **: 22:27:44.936: invalid cast from 'wxPizza' to 'GtkCellLayout'
2025-08-19 20:27:44.936201+00:002025-08-19T20:27:44.936201790Z
2025-08-19 20:27:44.936218+00:00(bambu-studio:326): Gtk-CRITICAL **: 22:27:44.936: gtk_cell_layout_get_cells: assertion 'GTK_IS_CELL_LAYOUT (cell_layout)' failed

[thelamer]

You are not running the latest container.

[Glujaz]

I did not know there was an update, and I also thought that I already did one just the day before. But my docker knowledge is not big...
I used the truenas GUI to execute the update just to be sure to update, and now...... I mean, it went from barely not useable to perfect !

So, thanks for mentioning the update. Issue is closed now :)