The cheapest PCs are ones that didn’t make the cut for Windows 11, so I’d look for 6th or 7th Gen. RAM is cheaper as well. Then you can build something cheap, and iterate from there.

For a beginner, check out Unraid. It’ll do 99% of what you want without being a hassle.

Check out refurbished Lenovo Thinkpad TinyPCs. Small, quiet and durable.

I’d always recommend not spending any money until you know what you really want and even then use any spares/old stuff you already have. The first build should be a true “budget build” :D

my i7 3700k with 8gb of ddr3 handles multiple 4k streams just fine. it doesn’t transcode on the fly super well, I transcode everything to hevc before putting it on the server. it will happily transcode a single 4k stream in the rare case a client doesn’t want hevc.

point is, don’t target latest gen simply because it’s latest. 12th gen is definitely newer than I’d opt for, and while 16gb is cool, 32 is absolutely overkill for the use cases you listed (other than the llm, which others properly list as “don’t worry about it unless you have 3k in your back pocket to shell on gpus”)

Just one thing that could be easy to forget. Power usage. Cost adds up as the hours and days pass :)

12th Gen is definitively neat, but overkill for what you are doing. 8th or 9th gen is fine. 16gigs is fine, 8gig would also work

I really like the Lenovo m720q line of mini pcs (used/refurb) but they will not fit an hdd.

/edit: I mean an 3,5” HDD. They will fit one 2,5” HDD fine.

For distro I’d recommend God’s Chosen Stock Debian. You want prioritize smooth upgrade paths from one major version to another since the server should last a long ass time, and Debian is that. Debian has the eyes on it to make it chill and if there are things to do they’re caught and noted (and more often something is set up to just do it for you).

So no, distro doesn’t matter as long as you can get your software going. But you’re not going to beat the insane stability of Debian. Any OS based on it is just an echo, which may even involve custom kernels which add more work for the specific distro maintainers and may slow critical security updates.

I run no GUI but you can def have one. I just SSH in. I do everything in containers (Docker 1st and I’m moving to Podman - but Podman is too heavy a lift to start, Docker compose files are always right there, use em) and LXC containers run by Incus (it has a web gui as well!).

I’ve been looking at “games on whales” https://games-on-whales.github.io/ to stream games from a beefy server to weaker computers and tablets. Might be a good route for your games without having to directly game on the server.

I have a 4 core 7th gen Intel CPU and it seems fine with doing x265 transcoding on multiple streams, but I’ve never tried finding out how many 4K ones it can do at once. I have 32 GB of RAM (I bought cheapo laptop RAM before the true AI shitshow, praise be, stuck them in with a DIMM board cooked up by China). I had 16 before that, and it was fine. But I just kept adding services.

An LLM will eat your RAM, you’d want 32 GB to be able to run a 16 GB one. But a GPU will be much, much faster. LLMs are just matrix math - that’s why they’ll never be able to think for real - and GPUs are sicko good at that kinda math (cause screens are a matrix!). I don’t fuck with LLMs though, no ethics in that whole shitshow.

Good luck!

Acronyms, initialisms, abbreviations, contractions, and other phrases which expand to something larger, that I’ve seen in this thread:

Fewer Letters	More Letters
AP	WiFi Access Point
DNS	Domain Name Service/System
IMAP	Internet Message Access Protocol for email
LXC	Linux Containers
NAS	Network-Attached Storage
PSU	Power Supply Unit
Plex	Brand of media server package
PoE	Power over Ethernet
SATA	Serial AT Attachment interface for mass storage
SMTP	Simple Mail Transfer Protocol
SSH	Secure Shell for remote terminal access
TLS	Transport Layer Security, supersedes SSL
VPN	Virtual Private Network
nginx	Popular HTTP server

---

[Thread #108 for this comm, first seen 21st Sep 2026, 05:00] [FAQ] [Full list] [Contact] [Source code]

random resource - lots of options. look up *ARR stacks

https://github.com/Rick45/quick-arr-Stack

LLMs

dont bother unless youre shelling out for a big ol gpu etc

Strongly recommend looking into Proxmox hypervisor as your base OS. Then install as many Linux VMs/LXCs as you like, with backups/snapshots all taken care for you. I’m on an Intel 8th gen CPU and have 3-4 VMs running, including Jellyfin, Home Assistant OS, an assortment of random docker images and a dedicated headless Debian VM for various AI projects. The CPU is more than enough, your 12th gen target will breeze through it. Get more RAM than you think you need - it will become the bottleneck for what you can do, not your CPU. I’m running 64GB and don’t regret it!

For video encoding/decoding like Jellyfin/Frigate, make sure you stick with Intel CPU for hardware codec support via QuickSync, otherwise you’ll need to add a GPU.

I can speak to this because one of my old servers was pretty much the machine you’re proposing.

Briefly, in terms of distros, choose what you like (within reason). I’ve always trended towards dietpi, which is extremely light. I actually just stood it up on the Raspberry Pi 4B last night. Headless and by itself it takes a whole whopping 300MB of RAM at idle. Even with all this running,

Synapse

PostgreSQL

Cinny

Arcane

qBittorrent

RDT-Client

Radarr

Sonarr

Prowlarr

Docker

Dropbear

Fail2ban

Tailscale

Syncthing

Argon One

TigerVNC

it sits under a GB

Do your own research, etc.

Pushing 4k streams, it’s not overkill at all, but by the sounds of it, you’re meaning to transcode on the fly? Be aware of AXV-512 support issues etc in Linux vs diff chips. Actually better make sure that your i5 has AXV-512 (not that it’s particularly required for transcoding but it helps) so double check.

Yes, you can run reasonable LLMs on that machine, though without a GPU, you’ll need to investigate exactly which backend runner best supports your CPU.

If you’re looking at a dense model on that rig, consider something in the 4 to 9 billion parameter range. Most comfortably, something like a Qwen3.5-4B, Gemma-4-E4B, MiniCPM or Tiny Ling…or a larger MoE (I like grug 35b-a3b, but Arcee Trinity mini or even GPT-OSS 20B could work) but the more parameters the slower it’ll run. OTOH, just more params =! always better. See -

https://reddit.com/comments/1w47xpd

Gaming wise, obviously you want something that can run Proton, wine, DOSBox, etc. I can’t claim that my pet distro runs all of those OOB because I’ve never tried, but it’s been my general experience that Ubuntu variants seem to do gaming best.

Anything slightly better than a potato will run all of that except the llm’s. To have a useful llm you need to spend circa £1k on a GPU and even then it won’t come close to Claude or chatgpt. Skip the llm’s for now. Just get yourself a solid little server to do everything else you need. Add dedicated hardware specifically for llm’s in future if you really want to. An old used gaming pc will be plenty for your server. I personally use Debian , but there are loads of suitable options available

Like all projects, you should think about needs, budget, and time. You’ve got a good list of what you want to run which is a good start. Budget is next, particularly considering the price of hardware these days.

FWIW, I run Jellyfin, Navidrome, the entire arr stack, Forgejo, Paperless, mealio, and Immich on a NAS with 24GB RAM and have never needed more. 8GB would have been a bit tight but probably doable. 16 will be plenty for you.

16GB ram is plenty for everything but getting into LLM’s.
If you do that then more is always better… 32GB plus a 6-8GB GPU would be a basic setup for qwen 35b-a3b for example

You can run an LLM, but it will be limited. Look into Open Vino Model Server.

I’d recommend a minimal install distro (Ubuntu server, Arch, Gentoo, etc…), then add the things you need from there.

Document every step. <— important