Raspberry pi5 + GPU

[reaster]

Hey all, first of all nice work to the devs,
i'm running a jellyfin instance serving in average 3-4 streams in continue to almost 35 friends and family,
for this i used an old workstation with a basic nvidia gpu on it, it worked fine, but it's an old intel cpu that hog way too much electricity, i would want to try a setup consisting of a raspberry pi5 with a gpu connected over the pci interface, i know this stuff can be sketchy, so, did any of you already managed to do a similar thing?
on paper it should work without issues,
thanks for your potential answers

[TheDreadPirate]

Definitely don't do this.

The CPU on the RPi and the narrow PCIe interface will severely hinder your GPU's transcoding performance.  Also remember that audio transcoding is performed on the CPU.  A RPi doesn't have enough CPU performance, memory bandwidth, or PCIe bandwidth to properly run a jellyfin server with media that requires any amount of transcoding.

The better option would be to get a bottom SKU, 12th gen and newer Intel CPU, like an i3 12100 or Intel N100/300.

[reaster]

hi, the raspberry pi5 cpu wouldn't be enough for simple audio transcode really? i'm running a i5-3350P on my main machine and i would expect same average of performance from my pi, is that delusional?
running os's without X11 save ton of cpu and ram

[TheDreadPirate]

Based on the benchmarks I am looking at, the 3350P is marginally faster than the RPi 5. But there is more to consider for a media server than raw compute. And equal benchmark scores don't tell the full story in terms of real world performance. There is still the issue of the single PCIe 2.0 lane available for the GPU. There are limited options for storage connectivity. Upgradeability is nonexistent.

For your user count, you need to ensure you have the headroom to handle both transcoding while still keeping the UI responsiveness high for others. And the RPi5 just doesn't have it.

12th gen and newer CPUs can be very efficient if you turn off turbo.

[reaster]

well i could agree with you on the single pci lane maybe that could cause issues, but on the side of ui responsiveness, i've tried to run a jellyfin container on my pi, and i could stream in direct play many 4k streams without issues, the only issue i got was remux and re encoding, but on the main system i'm running a nvidia 960 with 2Go of vram, and using the system of jellyfin that encode a chunk of video, then encode another chunk (instead of encoding the whole video) i can overcome the vram limitation and handle maybe 4 4k streams re encoded in h264,
for the pi5, using pci 3.0 i think 8Go/s of speed on the lane could be enough?

well maybe i would be the first the test it

[TheDreadPirate]

The RPi 5 has a single PCIe 2.0 lane, not 3.0.

https://datasheets.raspberrypi.com/rpi5/...-brief.pdf

Page 3.

It's up to you. I've voiced my opinion.