Hardware Acceleration with Oracle Cloud Ampere Processor

[fogolin]

Hi there! I'm quite new to home labs and personal servers, and started running a server on Oracle Cloud with their free tier to try out a few different services, including Jellyfin, Code Server, n8n and a few others. Oracle offers a very generous free tier for cloud computing:

Arm Compute Instance
Arm-based Ampere A1 cores and 24 GB of memory usable as 1 VM or up to 4 VMs.
3,000 OCPU hours and 18,000 GB hours per month

I'm running it using Ubuntu 22.04 and Docker, so my Jellyfin installation is using the latest image of linuxserver.io (which btw is an official distribution by Jellyfin's documentation), running version 10.8.13. The system works great, I can easily manage libraries, run plugins and more, but couldn't manage to make Hardware Acceleration work, and depending on the device I'm using, if the server has to manage transcoding via software only, it can take up the whole processing power of all 4 cores to manage a single transmission, and that's not great.

With Raspberry 5 and many new computers (including Apple's) using ARM processors, I was wondering if there's something we can do to manage it in the architecture, seems like the future lies on ARM for many (if not all) home use cases.

So I've checked the documentation on Jellyfin to see if they shed any light on the matter, but despite saying which types of supported hardware acceleration are available, they don't provide much more information in each of those... Looking on the old Reddit forum I found this thread, in which they discuss exactly this issue.

Running this simple command I can see that there is in fact a GPU on this machine:

Code:

root@oci:/home/ubuntu# lspci -nn | grep -Ei "3d|display|vga"
00:01.0 Display controller [0380]: Red Hat, Inc. Virtio GPU [1af4:1050] (rev 01)
root@oci:/home/ubuntu# lshw -C display
  *-display               
      description: Display controller
      product: Virtio GPU
      vendor: Red Hat, Inc.
      physical id: 1
      bus info: pci@0000:00:01.0
      logical name: /dev/fb0
      version: 01
      width: 64 bits
      clock: 33MHz
      capabilities: msix bus_master cap_list fb
      configuration: depth=32 driver=virtio-pci latency=0 mode=1024x768 visual=truecolor xres=1024 yres=768
      resources: iomemory:800-7ff irq:46 memory:13019000-13019fff memory:8030000000-8030003fff

Thus, as there is a GPU, I've mounted the device on my Docker container, and it shows up on my server setup:
   

But despite it seemingly working, when I go to the logs, to check out if it really worked, it tells me it is still using regular transcoding:

Code:

Stream mapping:
  Stream #0:0 -> #0:0 (h264 (native) -> h264 (libx264))
  Stream #0:1 -> #0:1 (copy)

My system runs on Ubuntu 22.04.3 LTS, so the latest available in LTS. Can anyone help on this? Is there a way to use hardware acceleration on Oracle Cloud? Do I need to change the transcoding options on my settings? Does anyone know if there is a branch or even a fork of the project that enables this? I'd appreciate your thought on the matter!

If I've posted it on the wrong place I'm sorry, please let me know. Hope to hear from anyone soon!

[Tim]

which btw is an official distribution by Jellyfin's documentation

Not correct. We do not consider LinuxServer.io an official distribution. Our documentation states the following:

Jellyfin distributes official container images on Docker Hub for multiple architectures. These images are based on Debian and built directly from the Jellyfin source code.

Additionally the LinuxServer.io (Dockerfile) project and hotio (Dockerfile) distribute images based on Ubuntu and the official Jellyfin Ubuntu binary packages.

[TheDreadPirate]

We do not officially support it. Hypothetically, you would use VAAPI or Video4Linux and as long as all the drivers for the Virtio GPU are installed it might work.

But you are in uncharted waters. You're on your own. I have no idea beyond what I suggested above as a starting point.

[fogolin]

I've tried it with both, VAAPI and Video4Linux but both resulted the same, not hardware transcoding. Thanks for your reponses and I apologize If I got the official distribution thing wrong.