Currently running all my docker compose containers on my gaming PC. 15 containers in total. Mostly *arr stack, plex, immich, home assistant, actual budget and jellyfin. Running on Mint.
Want to get these onto a dedicated pc. I have a mini with a I5 10-600, 32GB ram. I’ve played with it a little with jellyfin, on Debian and don’t think I was able to get quick sync enabled with my testing, and one transcode pretty much maxed out the CPU usage. To use this PC, I’d need to buy a 4 bay USB HDD enclosure.
So, basically I’m just wondering before I spend money and time if the hardware is even capable enough for my usage. 3 concurrent streams is probably the most it’d ever see, ideally with no more than 2 transcodes. Immich, home assistant etc are all pretty new and just in testing for now, but would only have 2 users total. Mostly using Plex, jellyfin is also in testing so it’ll be ready if plex enshitifies too much.
Acronyms, initialisms, abbreviations, contractions, and other phrases which expand to something larger, that I’ve seen in this thread:
Fewer Letters More Letters LXC Linux Containers NAS Network-Attached Storage NUC Next Unit of Computing brand of Intel small computers RAID Redundant Array of Independent Disks for mass storage
[Thread #16 for this comm, first seen 18th Jun 2026, 00:10] [FAQ] [Full list] [Contact] [Source code]
I’d call it decidedly overpowered for what you need - though I put my storage elsewhere, I’m using an 8th gen intel i5 for that and more.
With QSV, you can handle 2 transcodes simultaneously easily. I mentioned recently on mine that I had 6 simultaneous, of those were 4 transcodes and 2 direct streams, and utilization was barely more than where it idles at (while running everything else).
I don’t know much about quicksync but it’s very likely that you’re just missing something on debian to have it working. I believe the hardware in that mini pc (particularly the 32gb of ram, that’s practically gold) is more than capable of running all the services you listed.
Figure I can work out stuff like quick sync easy enough. Main concern was spending $100 or so on the USB enclosure and hours setting it up just to have to go back to the more powerful PC.
If you’re going to run the drives as individual drives in the enclosure, you should be fine but if you set them up as a RAID array, USB won’t work well. The connection is too unreliable and will cause issues. I’ve tried with a 10-bay USB-C enclosure using unRAID, Fedora, and Debian. I tried multiple cables. It just kept dropping the connection during large transfers.
Good to know. Storage space is expensive though and anything that’s using the big drives can be downloaded again if needed. So plan was to use online backup for the stuff I can’t lose like immich photos, actual budget etc and if a media drive crashes, I’ll just redownload.
I wonder if it was just a shitty chipset, or if it was overheating.
Nah, I looked into it and USB is just super unreliable for any sort of array. The drives all worked just fine if I ran them as individual drives.
Yeah but like… where is the weak link? Is there some deficiency in the protocol itself? Or the implementation in the chipset on one or both sides? Or bandwidth, or overheating?
In my case, it wasn’t overheating. Drives rarely got above 75-80F. There were two large fans and plenty of ventilation to pull air through the enclosure. From what I’ve read, mostly on unRAID forums, it is USB that is the issue.
I don’t run the 'arr stack, but for comparison, one of my servers is a Optiplex 7020 SFF / i7-4790 / 32GB RAM. It’s running around 45 containers now with ease from automatons with n8n to streaming audio with Navidrome. In fact, it spends most of it’s life in an idle state with load averages looking like .20/.27/.32. I don’t do any transcoding tho, so I can’t speak to that as far as resources. I see others have made suggestions in that dept. As far as the 4 bay USB HDD, you’re going to need some storage space at some point or another so it’s not like it would be wasted effort.
I run Jellyfin with hardware transcoding via QuickSync (through VAAPI) on a 10th-gen i3 NUC, 16gb RAM, and it seems to handle it comfortably. Your hardware should definitely be capable.
This is Proxmox with Jellyfin as native install in an LXC so it’s not exactly identical to your current setup, but what I found was that passing the device through is only half of it - the host also needs the Intel driver in place for QuickSync to actually work. Try running vainfo on the host to check, if it doesn’t list H.264/Hevc encode, the GPU block isn’t available yet and Jellyfin will fall back to software - this is probably what you’re seeing with the maxed out CPU.
After THAT, make sure Jellyfin has permission to use the passed-through device.
I was just going through this last week, migrating JF off my NAS, and it took some fiddling to get it working.
Did you pass the igpu to container?
Yes, but probably not correctly. That’s pretty much when I called it a day and haven’t tried to use it since.
Probably not helping. All i did was passing /dev/dri as device when i setup my jellyfin container . I
Did you check the render group matches? https://jellyfin.org/docs/general/post-install/transcoding/hardware-acceleration/intel#configure-with-linux-virtualization
The most useful part of those articles is the ffmpeg command you can run inside the container to test it.
I didn’t. Its docker container running on unraid fwiw.
I don’t think I did as that doesn’t sound familiar. Something to look into. Thanks.
@Itsamelemmy
You need to pass your gpu to the container.
Check the jellyfin docs for your specific gpu, but por my amd it’s some like thisdevices: - /dev/dri/renderD128:/dev/dri/renderD128Olso check on jellyfin docs for needed pkgs for hard accel to work on the host
I remember putting something like that into the compose file, but didn’t spend a lot of time trying to figure out why it didn’t work.
Everyone will screech but you should just run openclaw or similar and have it go through your logs. It’ll tell you the exact name of your igpu device and help you figure out why it’s not working. It’ll also help manage your queues, quality indexes, space, etc.
It’s reduced my server maintenance by at least 2/3rds.
Yeah. Let’s not learn what the logs means ourselves or anything.
Jfc, no one wants to think for themselves anymore. Your brain is shrinking. Read a book.
Parsing logs and pointing out anomalies isn’t solving anything. Second, your ass was just going to Google it anyway and end up the the same documentation. You’d just take longer with more effort to get there 🤷♂️
You’d just take longer with more effort to get there
That may be. But I’ll retain what I learned. And the effort IS THE POINT!
I’ll repeat myself: GO LEARN IT YOURSELF.
What value does rote memorization of a command, that varies by the system you are on, bring?
