Intel A380 HWA & other general questions regaridng Jellyfin

[TechZen]

Greetings fellow Jellyfin Connoisseurs,
As I've been working on setting up my new dedicated Jellfyin server, I've noticed some significant stutters, and down-right complete failure to play media if it's being trans coded.

I am wondering if I can have someone more experience, help share their insights into why how/why my server may be struggling in this regard. I admittedly am very much a newb, however I was expecting better results, especially considering my server hardware setup (see below).

As you can see in the attached screenshots, my Intel A380 GPU seems to be doing QSV HWA, which is good! However, I am concerned with the following:

1. Is it typical for the dGPU to be hitting 100% utilization for transcoding a single 4k stream?
   
2. Is my HWA config seem correct (see attached)? I've double checked it against the HWA docs, and don't see any glaring issues (all other settings are default)
   
3. Is there a best practice for what a client's bitrate should be? It seems to prevent transcoding I must have the quality setting to have a bit-rate higher than the media source.
   
4. What is a rough estimate on the number of concurrent 4k streams I should be able to handle if transcoding, direct streaming, or a mix of the two.
   

Jellyfin Install:
Jellyfin Server Windows x64, ver:10.8.13

Jellyfin Client(s): 

• LG C1 OLED (via WebOS App)
  
• iPhone 14 Pro (native iOS App)
  
• Desktop/Laptops using Windows Thick Client
  
• Additional devices using the HTML web client.
  
• Xbox Series X via the Edge Browser (as it seems the native Xbox app was deprecated)
  

Hardware:OS: Windows 11 Home 23H2
CPU: Intel i5-12400 with Intel 730 iGPU
GPU: Intel Arc380 w/ 6GB of VRAM, driver ver: 31.0.101.4575
RAM: 2x8 GB DDR4 @ 3200 MHz
Storage: 512 GB M.2 SSD (Used for Server/Cache Install)
Storage #2: 12 TB Seagate Exos 7200 RPM Hard Drive

Media:
All my media is ripped 4k Blu-rays using MakeMKV. I am not using handbrake to transcode my media to H.264, as I want to maintain the highest audio/video quality possible (however open to suggestions here).

Thanks for your time, insights, and help!

Sincerely,
TechZen

[TheDreadPirate]

As you can see in the attached screenshots, my Intel A380 GPU seems to be doing QSV HWA, which is good! However, I am concerned with the following:

1. Is it typical for the dGPU to be hitting 100% utilization for transcoding a single 4k stream?
   
2. Is my HWA config seem correct (see attached)? I've double checked it against the HWA docs, and don't see any glaring issues (all other settings are default)
   
3. Is there a best practice for what a client's bitrate should be? It seems to prevent transcoding I must have the quality setting to have a bit-rate higher than the media source.
   
4. What is a rough estimate on the number of concurrent 4k streams I should be able to handle if transcoding, direct streaming, or a mix of the two.
   

1.  Yes.  ffmpeg will always utilize your GPU as much as it can.  Like a slow PC and a fast PC running the same game.  Both will be at 100% but one is rendering the game as 30 FPS and the other 300 FPS.

2.  MPEG2 and VP9 can be checked.  Otherwise it's fine.

3.  The client side bit rate is only there if they, locally, have a bad connection or need to transcode for some other reason.  Otherwise just max it out on the client side and on the server side you can set a per-stream limit.  Both globally and per-user.  Only really needed if you have a limited upload speed.

4.  For plain transcoding (no tone mapping), my testing was around 20 streams, possibly a few more.  Tone mapping is VRAM intensive so a GPU with more VRAM can handle more streams.  For the Arc A380 6-8 tone mapped streams is what you can expect to achieve.  And that is if all of them are tone mapped streams.  I haven't tried a mixed situation, but I don't think tone mapping has too dramatic of an impact on how many non tone mapped streams your A380 can output.

IMO, for your LG TV, you should plan on buying an Android TV dongle.  You will have a MUCH better experience using Jellyfin than with the WebOS client.  Also, based on another C1 user's discussion in the Matrix chat, the C1 is already several WebOS versions behind the latest.  Another reason to get a cheap Android TV dongle.

As for your media and storage, 12TB is much less storage than you think it is if you don't recompress your rips.  Since you have an A380 you can encode to AV1.  I don't recall if Handbrake supports QSV AV1 (I use ffmpeg).  I achieve about a 70-80% reduction in my 4K HDR content when doing this.  A bit more once I started stripping out the audio and subtitle tracks for languages I don't need. And there is minimal quality loss.
Another consideration for re-compressing your rips is if you have a limited upload speed. You are more likely to not have to transcode your media for remote clients when the source file has a lower bit rate.

[TechZen]

As you can see in the attached screenshots, my Intel A380 GPU seems to be doing QSV HWA, which is good! However, I am concerned with the following:

1. Is it typical for the dGPU to be hitting 100% utilization for transcoding a single 4k stream?
   
2. Is my HWA config seem correct (see attached)? I've double checked it against the HWA docs, and don't see any glaring issues (all other settings are default)
   
3. Is there a best practice for what a client's bitrate should be? It seems to prevent transcoding I must have the quality setting to have a bit-rate higher than the media source.
   
4. What is a rough estimate on the number of concurrent 4k streams I should be able to handle if transcoding, direct streaming, or a mix of the two.
   

1.  Yes.  ffmpeg will always utilize your GPU as much as it can.  Like a slow PC and a fast PC running the same game.  Both will be at 100% but one is rendering the game as 30 FPS and the other 300 FPS.

2.  MPEG2 and VP9 can be checked.  Otherwise it's fine.

3.  The client side bit rate is only there if they, locally, have a bad connection or need to transcode for some other reason.  Otherwise just max it out on the client side and on the server side you can set a per-stream limit.  Both globally and per-user.  Only really needed if you have a limited upload speed.

4.  For plain transcoding (no tone mapping), my testing was around 20 streams, possibly a few more.  Tone mapping is VRAM intensive so a GPU with more VRAM can handle more streams.  For the Arc A380 6-8 tone mapped streams is what you can expect to achieve.  And that is if all of them are tone mapped streams.  I haven't tried a mixed situation, but I don't think tone mapping has too dramatic of an impact on how many non tone mapped streams your A380 can output.

IMO, for your LG TV, you should plan on buying an Android TV dongle.  You will have a MUCH better experience using Jellyfin than with the WebOS client.  Also, based on another C1 user's discussion in the Matrix chat, the C1 is already several WebOS versions behind the latest.  Another reason to get a cheap Android TV dongle.

As for your media and storage, 12TB is much less storage than you think it is if you don't recompress your rips.  Since you have an A380 you can encode to AV1.  I don't recall if Handbrake supports QSV AV1 (I use ffmpeg).  I achieve about a 70-80% reduction in my 4K HDR content when doing this.  A bit more once I started stripping out the audio and subtitle tracks for languages I don't need.  And there is minimal quality loss.
Another consideration for re-compressing your rips is if you have a limited upload speed.  You are more likely to not have to transcode your media for remote clients when the source file has a lower bit rate.

Sir Dread, you're the best damn pirate I've ever seen. Really appreciate the time to respond back with all the advice!

One quick question, in your experience would I see a huge benefit of re imaging and running Linux Ubuntu bare metal, rather than W11?

Thanks again!

[TheDreadPirate]

Most of the benefits of Linux over desktop Windows are related to uptime. Like updates not forcing reboots, there aren't any power saving features that might spin down the disk or turn off the NIC, or just straight up put the computer to sleep. Those things can be turned off in Windows but I've seen, personally and anecdotally from other users, where big Windows updates or driver updates turn some of those back on.

There are also performance improvements with switching to Linux, but it is not as dramatic of a difference with higher end hardware. Intel Atoms, for example, hugely benefit from running Linux instead of Windows.

If you don't already know how to use Linux, you should just plan to run Jellyfin on Windows for now. Maybe setup a little VM or something to familiarize yourself with the Linux command line before making the switch for your Jellyfin host.