Broken Gemini trying to Passthrough Nvidia GT710 on Jellyfin

[Polos123]

I bought into the hype that setting up a server was easy. I have little to no knowledge of programming languages but with AI and YT, I bit the bullet and jumped in.
I have a
Ryzen 5 2600
Gigabyte B550M DS3H AC with latest bios
32gb ram
Asus Nvidia GeForce GT-710 with GK208B.chipset to give out an HDMI
I'm running the latest Proxmox 8.4.1
and have Jellyfin in a container.
I have spent the last three days, of 16 hours a day with Gemini AI trying to passthrough this graphics card to Jellyfin so I can use the bit of on card transcoding. I cannot get the card to appear as a PCI on Resources in the Jellyfin container. This is what I have tried and this is what Gemini is telling me to approach the forum with:

When posting on the forums, provide them with:

• Your Proxmox VE version (pveversion -v).
• Details about your server hardware (Motherboard, CPU).
• The exact model of your NVIDIA GeForce GT 710.
• A clear description of your goal (passing the GPU to an LXC for Jellyfin transcoding).
• The steps you have already taken (mention repository cleanup, attempts to blacklist nouveau, attempts to purge and reinstall the 470 driver, disabling Secure Boot).
• The outputs of the key diagnostic commands we used:
  • lspci -vnnk (the full output, not just the grep)
  • dmesg | grep -i nvidia (the full output)
  • dmesg | grep -i nouveau (the full output)
  • cat /etc/modprobe.d/blacklist-nouveau.conf
  • cat /etc/modprobe.d/vfio.conf (to show it's commented out)
  • cat /etc/modules (to show vfio is commented out)
  • dpkg -l | grep nvidia

Explaining the persistent Nouveau conflict despite your blacklisting efforts and the lack of kernel driver output from lspci will be key for them to understand the unusual state of your system.

Can Anyone help???? Two versions of Gemini AI have left the building with their heads hanging in shame. I want a server where I can watch films on Jellyfin and Plex, witht he GPU doing most of the transcoding.

I knew I would get in bother starting this adventure.

 

[dcsapak]

if you want the card to use for transcoding, you have to install the correct nvidia driver on the host, you can use e.g. this debian guide: https://wiki.debian.org/NvidiaGraphicsDrivers

if you have successfully installed that, there are a few devices you have to give to the jellyfin container:

/dev/nvidia-caps
/dev/nvidia0
/dev/nvidiactl
/dev/nvidia-modeset
/dev/nvidia-uvm
/dev/nvidia-uvm-tools

(list taken from https://jellyfin.org/docs/general/post-install/transcoding/hardware-acceleration/nvidia )

you can give the container access to those devices with the gui under resources -> add -> device passthrough
or with the cli with

pct set ID -devX <device>

(replace devX with e.g. dev0,dev1, etc. and <device> with the path)

check out the link i posted above for e.g. permissions/users etc those devices need

in general I'd advise against using LLM chatbots to ask for such information, just because it will most often not give you an exact correct answer...

 

[Polos123]

Some old, same old: I keep getting

NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running.

Also the info you sent has me installing Version 535.183.01 drivers, yet even Proxmox is telling me to install the nvidia-tesla-470-driver.

And:
root@pve:~# cat /sys/module/nvidia_drm/parameters/modeset
cat: /sys/module/nvidia_drm/parameters/modeset: No such file or directory

I have followed the instructions in your first hyper link, but:

Make a dracut configuration file /etc/dracut.conf.d/10-nvidia.conf (you can actually name it anything you like, as long as it ends in .conf) with the following contents

Comes back with directory not found. I can, and have made the file /etc/dracut.conf.d and input the required parameters, but have no idea how to get to and include the "/10-nvidia.conf" portion of that instruction. being new to coding is really dragging this project into the mud.

It is supposedly possible to use this GPU, and I am very hesitant to get anything else until I can get this working.

Thanks for your help though.

 

[dcsapak]

Also the info you sent has me installing Version 535.183.01 drivers, yet even Proxmox is telling me to install the nvidia-tesla-470-driver.

no the link says install the 470-tesla variant, since it's listed on this page for the 470 driver: https://us.download.nvidia.com/XFree86/Linux-x86_64/535.183.01/README/supportedchips.html

did you use the nvidia-detect script? what was it's output?

also

Make a dracut configuration file /etc/dracut.conf.d/10-nvidia.conf (you can actually name it anything you like, as long as it ends in .conf) with the following contents

Comes back with directory not found. I can, and have made the file /etc/dracut.conf.d and input the required parameters, but have no idea how to get to and include the "/10-nvidia.conf" portion of that instruction. being new to coding is really dragging this project into the mud.

this is also only necessary if you use dracut which i doubt you do...

 

[Polos123]

Sorry for taking so long to get back to you, but i have been trying every permutation I can to get jellyfin to transcode on the GT710.
I am new to Linux so found a great guide which took me through the installation of a 470.run driver, but jellyfin keeps stopping me from playing files when the transcode the transcode Nvidia NVENC is selected. The method I followed lead me through the Nvidia site where *.run files are used for the drivers. The site you listed are *.deb and I have no idea how to switch between the two.
The dev setup you advised corelated with the set up I followed. I can get the nvidia-smi working on both my pve and lxc, but I have no idea how to install the .deb files with the method that got me so far.
I am on my fifth Proxmox rebuild after experimenting and messing up files and *.conf folders etc.

So in summary, I can install 470 drivers from Nvidia with the .run extension, but I cannot follow the only instructions I have made any progress with, using the 470-tesla.deb files in the links you kindly provided.

I know I am very close now, and it is so frustrating having to rebuild every time I make a series of mistakes to ensure I am not compounding problems.

II am picking up Linux on the hoof though, albeit frustratingly slow.