SOLVED: Trancoding performance suddenly poor

[yellowjacket_ur]

I noticed this afternoon that the transcoding performance of Jellyfin is far worse than ever before. I am running Jellyfin in an Unraid Docker with an Arc A380 passed into the container for transcoding using VAAPI. When I trancode I see activity on the card. I've used the card for transcoding before and performance has been good and I've never experienced this issue before. I was even able to transcode and stream HDR content remotely using Tailscale without the transcoding fps being an issue. I didn't consciously change any settings prior to this happening, but I did update my Tailscale plugin on my Unraid server.


FFmpeg log:
https://pastebin.com/VtzEfkfC

[TheDreadPirate]

Have you been transcoding before or has it been direct playing? The screenshots says the transcode reason is due to the bit rate exceeding some limit.

AFAIK, unRAID doesn't properly support Arc since it is still on kernel 6.1.X. Arc needs 6.2+.

[yellowjacket_ur]

I have transcoded in the past, as I am on the Unraid 7 beta and that adds support for Arc GPUs. There are no issues using direct play for any of my media, but this trancoding issue is new as of today. As recently as this past week, I transcoded my media while I was away from home to stream via Tailscale and the issue was not present.

[TheDreadPirate]

Do you know what kernel the beta is using?

[yellowjacket_ur]

I think it's using 6.8.12 per the release notes in my dashboard.

[TheDreadPirate]

Starting with kernel 6.8.5 there is a kernel driver bug that either breaks transcoding on Arc or tanks performance, depending on the ffmpeg command. You'd need to revert to 6.8.4 or jump ahead to 6.9.4

Not sure how you'd go about that with unRAID.

Did the performance issue start when you started using the unRAID beta?

[yellowjacket_ur]

No, I have been running the Unraid beta since the start of July and have since updated to the 7.0.0-beta.2 2024-07-12 version without any problems. Like I said, I was transcoding just fine this past week. The only things that I can think of that have changed since right before the issue are that I updated my Tailscale plugin and I manually enacted my mover because my cache was getting a bit full this morning. My transcode folder is in my media folder and should be kept on the cache drives but maybe it got moved to my hard drives?

It's honestly very strange, so I'm going to try restarting my server and see if that fixes it.

Update:
Restarting the server did not fix the issue.

[Fate]

No, I have been running the Unraid beta since the start of July and have since updated to the 7.0.0-beta.2 2024-07-12 version without any problems. Like I said, I was transcoding just fine this past week. The only things that I can think of that have changed since right before the issue are that I updated my Tailscale plugin and I manually enacted my mover because my cache was getting a bit full this morning. My transcode folder is in my media folder and should be kept on the cache drives but maybe it got moved to my hard drives?

It's honestly very strange, so I'm going to try restarting my server and see if that fixes it.

Update:
Restarting the server did not fix the issue.

Can you post a screenshot of the transcoding tab in jellyfin?

Can you run vainfo within jellyfin container?

[yellowjacket_ur]

Here are screenshots of the Transcoding tab.

When I try to run the vainfo command in the console under my Jellyfin docker in Unraid it gives the following error: bash: vainfo: command not found

Previously, I also updated the container to 10.9.10, but that didn't solve the issue, so I deleted that and reinstalled the 10.9.8 version. The older version has the same issue still.

[Fate]

Hm I'm not an Intel ARC expert, but I would expect selecting QSV whould be better supported than pure VAAPI.

Also you sure your trancoding folder /cache/Transcodes is on reasonable fast storage?

Also there should be no differences in transcoding between any 10.9.x versions so always go for the latest