GPU Transcoding on Unraid

[ClumsyMinty]

Some context for this: I couldn't get GPU hardware transcoding working on my jellyfin container installed on Unraid. I really struggled with the clarity of the official guides so I decided to just post the solution to my problem exactly in the way I understand it and hope others will too.

Essentially what I had to do was modify my jellyfin container by: adding this line to my extra parameters under Advanced View
Extra Parameters: -e NVIDIA_DRIVER_CAPABILITIES=all  -e NVIDIA_VISIBLE_DEVICES=all  --gpus all
That's the exact line I used, I copy and pasted exactly that, obviously without the Extra Parameters part.
Than I created 2 variables in the container editor.

Variable 1:
Name: NVIDIA_VISIBLE_DEVICES
Key: NVIDIA_VISIBLE_DEVICES
value: all

Variable 2:
Name: NVIDIA_DRIVER_CAPABILITIES
Key: NVIDIA_DRIVER_CAPABILITIES
Value: all

Aside from this you need to make sure NVIDIA drivers and NVIDIA container toolkit is installed. On Unraid, you can install the "Nvidia Driver" app/plugin from the built-in apps installer, the author at time of writing is 'ich777'.
The container toolkit just works, I assume it comes with the drivers.
I also installed "GPU Statistics" I don't think it's necessary but it's nice to see how the GPU is performing.
After that I just setup the hardware transcoding on jellyfin and was good to go.
There's 2 ways to verify the functionality. First, you can turn NVENC encoding on in Jellyfin and try to stream a show. The second way is run this command in the Unraid terminal: docker exec -it jellyfin nvidia-smi
If the command outputs a table with GPU statistics you know it worked.

[Potato Undercover]

Also, if you are ARC struggler, I just followed https://www.reddit.com/r/unRAID/comments...ding_with/ guide today, and it is working with the latest kernel fix from thor2002. It was extremely easy to set up, as opposed to what I tried earlier with having to tinker with kernel files(it failed). Only issue is that VPP tone mapping works, the other one makes HDR content fail to play. Both QSV and VAAPI work, not sure which is better. on unraid version 6.12.14

[Rafael Silva]

I subscribed just to thank you for this, you're awesome!

[Djondan]

Also subscribed to thank you and add for other idiots such us myself, rebooting unraid might be necessery after installing nvidia drivers. Spent a frigging hour on that

[alike]

Hi. I did this except for one perhaps major difference:

I used the official Limetech version of the Nvidia Driver.
And then I chose to install v570.133.07 since this was "patchable" according to this software https://github.com/keylase/nvidia-patch

Then I patched it according to the user script:

Code:

#nVidia Patch for Unlimited Streams
#https://github.com/keylase/nvidia-patch
wget https://raw.githubusercontent.com/keylase/nvidia-patch/master/patch.sh
chmod +x patch.sh
./patch.sh

It confirmed the first time I run the script



So it should be working.

Also, I can see that its working in the driver. Not sure what could be wrong. I used the Jellyfin parameters you shared with us. And inside of Jellyfin server I used the same settings I had on the same card in a windows 11 pro setup that has been working great.



Edit: Log report
https://privatebin.net/?1abc3a785948801f...No1Bop5RuQ

Edit2: Nevermind, i tried again without the HW acceleration and had the same issue. So I need to check my install perhaps.

Edit3: I found out the problem, I lost the files it was trying to access because my RAR2FS setup failed. I will report back when I have fixed that, quite sure the hardware transcoding is working like it should now.

Edit4: Yup, I got rar2fs to work again and now playback and HW transcoding is working. Thanks for the great guide.

[greebo]

Huge thanks, really appreciate that you posted this! Poor/out of date instructions were crushing me. Cheers 🇨🇦🇳🇿

[PET]

I have also just created an account to say thanks. I added : -e NVIDIA_DRIVER_CAPABILITIES=all -e NVIDIA_VISIBLE_DEVICES=all --gpus all
to the parameters and this made the encoding work. I had problem with a set of videos that were fairly old.