Hardware upgrade dilemma

[oopsmybad]

Because I have a Synology NAS, I've been stuck running 10.8.11 for hardware transcoding. It's finally time to upgrade, but I'm not sure which direction I should go, specifically due to subtitles.

We have bedroom TVs, and in the interest of kindness, turn the volume way down and subtitles on when we're going to sleep. My brand new TCL TV runs Google TV. If I use Exoplayer with the Jellyfin app, the PGS subtitles work - but the other type have to be burned in, and that fails because of my current hardware limitions. If I use libVLC, all the subtitles play - but with 4K HDR movies, when the PGS subtitle changes, it causes momentary frame jitter, which is annyoing. And libVLC doesn't support Dolby Vision.

So, I think I have two options:

1) Buy hardware that can properly support burning subtitles, and keep using Exoplayer on the TCL's Google TV. I'm not sure what specifically that hardware would be. Would a 10th gen I5 with a UHD 620 suffice? It would have to be capable of burning subtitiles into 4K HDR/Dolby Vision movies.

--or--

2) Buy a cheaper N100 device (like a $159 Beelink), and upgrade the streaming players in the bedrooms to using $90 Roku Ultras. In the living room and family room, I have 2021 and 2022 model-year TCL Roku TVs, and they play all subtitle formats natively... so I assume the current stand-alone Roku Ultras would also work.

What do you think my best option is? Thanks in advance for any and all opinions!

[Efficient_Good_5784]

Just as a heads up, the current version of Android TV dropped LibVLC and you need to have the latest release for ATV if you're going to upgrade your server to v10.9.
If you're curious about why LibVLC was removed, it was due to security issues and also that it was hard to maintain for the app.

If subtitles do not work on ATV, they should be burned into the video. Most to all modern GPUs (including modern iGPUs found in CPUs) can handle transcoding a couple of 4K streams well (the burning-in work part takes place on the CPU however). The issue will be that you're trying to burn them in on HDR/Dobly Vision content, so you need to make sure tone mapping is set up well in the transcoding settings.

I would offer a third option. You can build a new server and install an Intel ARC GPU. They're really good for transcoding content and you can find some for relatively cheap prices.

[oopsmybad]

I would offer a third option. You can build a new server and install an Intel ARC GPU. They're really good for transcoding content and you can find some for relatively cheap prices.

After doing some reading and research, I am falling in love with that option. I do have a few follow-up questions, if you have a moment.

With an Arc graphics card, are any operations still CPU bound, like burning subtitles? Or are all of those GPU bound? For cost reasons, if I can get away with buying a 12th gen I3, that would save some money.

Also, is it okay to buy an 'F' processor variant, one without the iGPU? It doesn't seem like Intel's Deep Link Hyperencode is destined for the Linux/docker version of jellyfin-ffmpeg anytime soon, and I think QuickSync still works, even without an iGPU, if you have an Arc GPU.

Finally, I think an A380 would be capable of burning subtitles into a 4k DV/HDR stream, transcode a 4K DV/HDR movie to 1080p SDR, and transcode 1080p/x265 to 1080p/x264 -- 3 total simulatenous streams. If I'm wrong about that, please let me know.

Thanks for your help!

[TheDreadPirate]

With an Arc graphics card, are any operations still CPU bound, like burning subtitles? Or are all of those GPU bound? For cost reasons, if I can get away with buying a 12th gen I3, that would save some money.

Burning in subtitles does partially involve the CPU and does slow down transcoding.  But getting a faster CPU, or CPU with more cores, won't speed up that process much/at all.  In my testing when subtitle burn-in is occuring, my CPU usage is about 20-30% spread across all 8 threads on my 12100.  BUT, that is 20-30% at 800Mhz.  The subtitle burn-in work load just doesn't trigger the CPU to boost.

Also, its the difference between transcoding at 1000FPS and 500-750FPS.  Still much much faster than required.

Audio transcoding also occurs on the CPU, which is purely single threaded.  My 12100 does not boost much for audio transcodes, indicating it is keeping up with the Arc GPU's video transcode.

Overall, I've yet to encounter a situation where my 12100 is bottlenecking my A380.

Also, is it okay to buy an 'F' processor variant, one without the iGPU? It doesn't seem like Intel's Deep Link Hyperencode is destined for the Linux/docker version of jellyfin-ffmpeg anytime soon, and I think QuickSync still works, even without an iGPU, if you have an Arc GPU.

Yes.  I had to turn off my iGPU, actually.  My main Jellyfin instance runs directly in Ubuntu.  The problem is that Intel Quick Sync does not allow you select which GPU it uses.  It will use the first available GPU, which is almost always the iGPU and not the Arc GPU.  If I was using docker (which I do for my test instances) I could specify a GPU to pass into the container.

So going with an F SKU CPU is fine.

Finally, I think an A380 would be capable of burning subtitles into a 4k DV/HDR stream, transcode a 4K DV/HDR movie to 1080p SDR, and transcode 1080p/x265 to 1080p/x264 -- 3 total simulatenous streams. If I'm wrong about that, please let me know.

Thanks for your help!

One thing to keep in mind is that Jellyfin currently cannot do HDR to HDR transcoding.  If the video needs to be transcoded for ANY reason, including burning in subtitles, it will tone map the HDR video to SDR.  Work is being done to add HDR to HDR transcoding, but there is no ETA on that.

With that out of the way, yes, it should be able to handle 3 streams, easily.

It should be able to handle at least 6, possibly more, 4K HDR to 4K/1080P SDR tone mapped transcodes. And I stopped my testing when I hit 20 simultaneous 1080P AV1/HEVC SDR to 1080P H264 transcodes.  I only have 4 users so that was more than enough.  

[Efficient_Good_5784]

As a general guideline, the more VRAM a GPU has access to, the better its tone mapping performance will be. So if you need to do lots of HDR to SDR conversion, go with an option that has more VRAM.

With an Arc graphics card, are any operations still CPU bound, like burning subtitles? Or are all of those GPU bound? For cost reasons, if I can get away with buying a 12th gen I3, that would save some money.

Some operations still are CPU only.
You are right with subtitles burn-in need to be processed on the CPU. This shouldn't be an issue on most modern CPUs (especially if the video transcode is being handled by the GPU instead).
Another common operation that's CPU only is audio transcodes, but audio transcoding is generally very lightweight, so most CPUs will not have any problem transcoding audio fast enough to not cause buffering.
An Intel 12th gen I3 sounds reasonable enough for your needs.

Also, in case you didn't know, QuickSync performance is nearly the same on all CPU models (when comparing the same generation of CPUs at least). The core count/tier level doesn't affect it as it's its own thing on the CPU.

Also, is it okay to buy an 'F' processor variant, one without the iGPU? It doesn't seem like Intel's Deep Link Hyperencode is destined for the Linux/docker version of jellyfin-ffmpeg anytime soon, and I think QuickSync still works, even without an iGPU, if you have an Arc GPU.

It's fine to go with an F variant, but if the price for getting a CPU with an iGPU isn't that much more, I would just get that instead. Having the iGPU would mean you have a backup in case the GPU ever fails.
Just know that you might have to disable the iGPU in the BIOS to make sure you're using the external GPU with Jellyfin in some cases.

And yeah, Intel Arc cards use QuickSync too.

Finally, I think an A380 would be capable of burning subtitles into a 4k DV/HDR stream, transcode a 4K DV/HDR movie to 1080p SDR, and transcode 1080p/x265 to 1080p/x264 -- 3 total simulatenous streams. If I'm wrong about that, please let me know.

As I started this post, more VRAM is important for tone mapping, so if you need more of that, get a model with more VRAM.
And like I mentioned about the CPUs, the Arc card's encode/decode performance for QuickSync are roughly the same for all models (of the same generation again), so you can get a cheaper Arc card and receive the same transcode performance.
But if you're happy with the price of an A380, that's a good option.

[oopsmybad]

Thank you for all the tips. After pricing out all my options, I bought a refurb (2 year allstate warranty) HP Victus w/ I5-13400, 16gig of ram and an A380 for $378. I always prefer to build a system myself with thoughtfully curated components, but the deal was too good to pass up, I think.

[oopsmybad]

I would offer a third option. You can build a new server and install an Intel ARC GPU. They're really good for transcoding content and you can find some for relatively cheap prices.

This was definitely the right solution. My new server can transcode 3 remux 4k/HDR -> 4K/SDR movies, one with burned in subtitles - but since I mount the Synology via NFS for the media, I fully saturate my gigabit ethernet connection, which is my next bottleneck.

Oh, and if anyone else buys one of these cheap refurb HP Victus 15L's, they're decent systems... but HP locks you out of advanced settings in BIOS, so you can't disable the UHD 730. This didn't bother me, as I planned to use docker anyway, and you can pass in just renderD129, which is the ARC380.

Thanks again!