Setting up a deticated GPU for transcoding / Hardware acceleration

[RobertUnger]

Hi!

FIrst.... I'm new to this TrueNAS - Jellyfin thing but i do have a little bit experience with Jellyfin on Synology DSM.

I've build a DIY NAS with TrueNAS as the OS and installed Jellyfin on it (NO Docker!). Now i want to use my dedicated GPU for trranscoding.

CPU: i5-13600
GPU: Intel ARC B580

I looked around the Internet to find some guides, glues or something that could point me in the right direction but found nothing that fits my needs. Evene from the docs Intel GPU Jellyfin i could'nt extract infos for my usecase. Maybe it's just my missing knowledge thats holding me back....

So here are some pictures from my settings....

Jellyfin settings:


Jellyfin APP config:


TrueNAS shell:


Doesn't looks like that my dedicated GPU is properly installed....


Everytime i start a UHD movie the CPU Usage ramps up. So there must something be wrong.

Please help! Thanks!

Ps: pleas feel free to move this post to the right forum if i missplaced it.

[alleycat]

Hi!

FIrst.... I'm new to this TrueNAS - Jellyfin thing but i do have a little bit experience with Jellyfin on Synology DSM.

I've build a DIY NAS with TrueNAS as the OS and installed Jellyfin on it (NO Docker!). Now i want to use my dedicated GPU for trranscoding.

CPU: i5-13600
GPU: Intel ARC B580

I don't think the kernel update for the driver are ready yet for TrueNAS.

https://forums.truenas.com/t/intel-arc-b...port/38945

[RobertUnger]

I got the B580 because the older gpu's are just a few bugs cheeper, so i bought the better one. Well, so i have to use the iGPU till they add support for the B580.

But the question is... is my setup configured the right way?
I mean.... CPU usage goes up but i dont know if it is Software transcoding or if it is the iGPU. Is there a way to find that out other than activating hardware acceleration and thinking it's working?

[ArneR]

Remember that audio is being trancoded using the CPU, so a spike in usage is to be expected, unless you have a device that can direct play the spesific audio format. 

You could post a log, but a quicker answer is to simply click on the gear icon in the lower right corner when playing a video, and selecting playback info, it will state there if it is direct playing or transcoding. 
you might have to lower the bitrate to force transcoding.

   

EDIT: If you have the option, you should return that B580 and opt for buying an older Arc A310/A380, they have the exact same media encoders in the silicon, the B580 is just a waste of money and energy.

Besides that, the Arc A series should have the low power encoder options set, the B580 should be no different one should think.

[RobertUnger]

Remember that audio is being trancoded using the CPU, so a spike in usage is to be expected, unless you have a device that can direct play the spesific audio format. 

You could post a log, but a quicker answer is to simply click on the gear icon in the lower right corner when playing a video, and selecting playback info, it will state there if it is direct playing or transcoding. 
you might have to lower the bitrate to force transcoding.

I know that Playback Info, and i can force to transcode if i use my phone. But i don't see if it's transcoding with the iGPU (Hardware) or with the CPU (Software). I don't have just a spike in CPU usage... the overall CPU usage is around 10% if i play a 4kUHD HDR with Atmos (Video quality in Jellyfin playback settings on my phone is set to 480p audio is set to auto, Bitrate of transcoding is 241 fps (10.5x) acording to the playback info).
I would like to have GPU stats i could watch like the CPU stats but i can't find that.

I guess, the question would be.... Spikes the CPU usage on the dashboard if the iGPU is used too or can i just see CPU usage if the CPU cores are used?

I know, that sounds a bit strange but as a windows guy i don't know a better way of telling if the GPU is used other then using a tool wich is showing me a graph and some numbers . 

EDIT: If you have the option, you should return that B580 and opt for buying an older Arc A310/A380, they have the exact same media encoders in the silicon, the B580 is just a waste of money and energy.

Besides that, the Arc A series should have the low power encoder options set, the B580 should be no different one should think.

Well, sending back the GPU is an option for shure. Is this 100% true, the B580 is no upgrade for transcoding?

[ArneR]

Go to your dashboard  and look at the "Ffmpeg.Transcode" log after playing a few minutes of the movie on your phone or whatever device you want, and look for "Stream mapping".
An example of transcoding of a HEVC 4K BluRay to h264 with QuickSync will look like this:

Stream mapping:
  Stream #0:0 -> #0:0 (hevc (native) -> h264 (h264_qsv))
  Stream #0:1 -> #0:1 (dts (dca) -> aac (libfdk_aac))

Notice the h264_qsv. If it was transcoding video with the cpu it would not say that. Nor would the performance be anywhere near what you mentioned I would think.
My guess is that it is using the iGPU, which is not slow by any means, especially your UHD770 with two media engines, as long as you're not trying to encode to HEVC at 4K HDR full bitrate.

EDIT: Forgot to answer the last part. There is zero benefit in going with Battlemage over Alchemist for QuickSync transcoding.
You can ssh into your truenas with ssh truenas_admin@truenas.local and run "sudo intel_gpu_top" while a movie is playing, it will then look something like this:

[RobertUnger]

Notice the h264_qsv. If it was transcoding video with the cpu it would not say that. Nor would the performance be anywhere near what you mentioned I would think.
My guess is that it is using the iGPU, which is not slow by any means, especially your UHD770 with two media engines, as long as you're not trying to encode to HEVC at 4K HDR full bitrate.

Thats the answer i was looking for. At least for the performance part.

especially your UHD770 with two media engines

This almost sounds to me like it even wouldn't make sense to do what @ArneR mentioned in his post and buy a A319/380

Let me ask this... in my house we have 3 People with 1 TV each. Lets say all of them would want to watch a 4k movie at the same time.....

What would i need to be able to realize this. Would the cpu be fine (i don't think so), would a A380 do the trick, do i need the B580 or is all of this to weak to stream 3 simultaneous streams?

Guys.... btw.... thank you very much for yor help 

[ArneR]

I... I'm a bit confused by this last answer, I was the one suggesting buying an A310/A380 over the B580. And I'm the same one that commented on the relative performance on your iGPU. There are not more than one of us (me).

The point still stands:
The B580 is not better than the A310.
An  A310 is better than the UHD770.
A UHD770 is probably more than enough for your usecase, especially if you limit the bitrate/resolution some.

Why don't you just test and see for yourself where the limits of the iGPU lies? But I think you will find that the issue lies in space constraints in the /transcodes folder by running multiple 4K streams at full bitrate at the same time. I'd suggest to enable throttling of transcodes and consequently deleting of old segments to save space.

I transcode to a temporary directory created in RAM on my server since I have so much of it, and after some testing I found that throttling after three minutes and deleting segments 5 minutes after they have been played works fine for me, I also have allocated 30GiB of RAM for this directory, since a single 4K HDR movie can use quite a lot unless throttled etc.

         

Open three tabs in a browser to your jellyfin server, and start playback of the same 4K movie in all tabs. What happens? Do they all play nice? If so, open another tab and start another. My guess is that it will crash at some point in time, but then at least you know the limits.

Your cpu should have no issues transcoding audio for your household.

[RobertUnger]

I... I'm a bit confused by this last answer, I was the one suggesting buying an A310/A380 over the B580. And I'm the same one that commented on the relative performance on your iGPU. There are not more than one of us (me).

Haha!..... i'm sorry, was writing in more forums... seems, i got confused with the names etc.

The B580 is not better than the A310.
An  A310 is better than the UHD770.

Ok, got it. So i have to test it. I'm still in the process of switching from plex. I guess i have to do it the next days. It's time anyways.

I transcode to a temporary directory created in RAM

I did the same thing. But i have only 32 GB RAM in total. I gues i could use 20GB or something. I'm only running jellyfin and pihole.....

I'll try out how much streams i can run...

[ArneR]

I did the same thing. But i have only 32 GB RAM in total. I gues i could use 20GB or something. I'm only running jellyfin and pihole.....

I'll try out how much streams i can run...

Weren't you running Truenas scale? Truenas use ZFS, which loves RAM for caching. 

Anyway, you can try setting a limit of say 16GiB for the transcode directory, and further down where you are to set the resource limits,set that limit to 20GB. Failure to set that second limit high enough will make the container crash when it runs out of memory. I found that out the hard way. 
After that you'll have to figure out how much RAM one 4K movie use by looking at the Truenas dashboard, the screenshots here are from transcoding a high bitrate 4K HDR movie to H264, with tonemapping on my Intel Arc A310:

     

   

Worth noting that Jellyfin at this time also was serving an episode of a TV-show:

   

RAM usage hovers around 9,1-9,5 GB. Without it doing anything it stays at 1,99 GB. So a reasonable amount of usage in my case is 7-7,5 GB per 5 minutes of transcoded video.
You probably need to invest in more RAM or set the throttle time and time before segments are deleted lower than my values I posted screenshot of before.

Also, note that the Arc A310 is capable of transcoding at 3.92x, which is adequate for almost 4 consequent streams of this quality level. Lowering the bitrate will naturally allow for more streams. This is in addition to the TV-show episode, so the limit is probably closer to 4x.

BTW. Regarding CPU usage, I saw a peak of around 15% in the beginning of the transcoding, mind you that is only transcoding audio from 7.1 TrueHD to stereo AAC since I was using Firefox on my pc to test. It then settled down to pulses around 5-7% when it needed to transcode the next segments.