Skip to main content


Now that Nvidia GPUs run on the Raspberry Pi, I want to see if they're good for Jellyfin transcoding (e.g. take your old 3060 and throw it on a Pi for a little #homelab media server...). jeffgeerling.com/blog/2025/ben…
in reply to Jeff Geerling

PCIe Gen2 x1 (500-ish MB/s) should be plenty - i see, you've massaged the config, Gen3 x1 - bandwidth won't be the issue.

If the driver runs on ARM, ffmpeg is popular enough to do so, as well.

I should actually read your blog post, now. 😅 ...one moment, please..

This entry was edited (2 days ago)
in reply to Multi Purr Puss

on my box - AMD Zen4 octa-core mini PC + OCuLink eGPU/Radeon 9060 XT - i don't get PCIe Rx/Tx stats in nvtop. (Manjaro, btw)

EDIT: i had to double-check, after the big December update of Manjaro - i'm still getting "N/A" for PCIe Rx/Tx, in nvtop. radeontop 1.4 doesn't even recognise my 9060, but that's off-topic. 😅

This entry was edited (2 days ago)
in reply to Multi Purr Puss

@platymew 90 series is a bit wonky with some features, I've had issues with the 9700 XT and R9700 both sometimes, depending on the day of the week it seems. With telemetry, that is.
in reply to Jeff Geerling

that'd be a 9070 XT, not 9700 XT. (Yes, there is an R9700 workstation class card) ..these product names!

Every single one should include a decoder wheel / slide rule! 🤣

in reply to Jeff Geerling

rasp pi fans they must have pi 6 lined up around the corner - one would think, i have my money down
Faster CPU, 32GB RAM, Wi-Fi 6E, NVMe support #compute per watt
in reply to Jeff Geerling

initially I was wondering about the neat GPU graphs, did they add them to #htop? Wouldn't be the first time that htop were surprising me with cool, new features. Checked my htop, couldn't find them though. Then read your blog post. And ah! It's a tool called nvtop. Good to know.
#htop
in reply to T_X

@T_X btop does have a GPU graph, but right now I can only get that working on x86 :(
@T_X
in reply to Jeff Geerling

"your old 3060" is still my current GPU. Can we roll this back to actual legacy GPUs?
in reply to 768kB 🚴☕🇬🇧🇺🇦🇵🇸

@768kb So far I don't have any 20 series cards that I can test on. I should pick one up!

I know 10 series and older aren't supported at all by the open kernel module :(

in reply to Jeff Geerling

since a PI5 is useless as Jellyfin server, I just went ahead and got myself a N150 box with 12GB RAM and 512GB NVME. In total it cost about same or less than just a 8GB PI5 inkl all accessories you need to run it (case, PSU, SD card or USB SSD). The N150 is a perfect homelab for me. It runs HomeAssistant, Jellyfin, TVheadend and other container and can transcoded 4 1080p streams in parallel. Super happy with it. I love my PIs for other things, but they just don't work for Jellyfin
in reply to DaAnda

@DaAnda I wouldn't say useless, but N100/150 is fine too. Some are getting way more expensive because of RAM shortage 😢
in reply to Jeff Geerling

useless in the sense that you can not really do transcoding. I ran JF on a RPI4 without transcoding, and it was "fine" for a while, but in the end it was mostly just a fancy library UI, unless you exclusively used KODI as client. Every other client basically relies on transcoding, and the PI5 unfortunately does not have suitable HW decoding and encoding support.
in reply to Jeff Geerling

- just out of curiosity, is any of this explained in repeatable steps people without "institutional knowledge" of this process can follow? By repeatable steps, I mean the list of cli commands in order.

Example
Step 1: execute "mkdir kernelfiles" //creates new directory
Step 2: "cd kernelfiles" //navigates into directory)
Step 3: sudo apt-get source XYZ --arg1 X --arg2 //downloads source
Step 4: execute "sudo apt install buildpackage1, buildpackage2" //installs neccesary tooling
Step 5: install kernel using ...

Not

Step 1: clone the kernel
Step 2: build the kernel
Step 3: some other vague descriptor

Would love to try it myself.

This entry was edited (2 days ago)
in reply to Jeff Geerling

how long until nvidia includes a compute module on the card?
The standalone GPU!
in reply to Jeff Geerling

"[…] take your old 3060 […]"
You live in a very different world than me. I am happy I got a 1060. And that's in my best computer.
Saying a 3060 is something people have laying around unused is a very different reality than mine.
in reply to txt.file

@txt_file I'm a bit jaded by working with the Nvidia drivers, which don't even support the 10 series anymore :(

I never bought an Nvidia card until the 30 series existed, but because of my YT stuff, I upgraded to 40 series last year on my streaming PC. I don't see a reason to go 50 series at all, though.

in reply to Jeff Geerling

if only nouveau would be able to fully support the hardware. But something something NV-RISCV and FALCON and probably more stuff.

Also sad AMD radeon/amdgpu is worse :-(

Thanks for your work. I am sad about the results but that's not your fault. ;-)