Good news selfhosted crowd! I pitched the idea of setting up a server, and she is down with the idea!

I want to run Linux and host things like Jellyfin, Audiobookshelf, Navidrome, maybe Plex (I currently host it off my Mac desktop but would move it to a dedicated server).

I know you can host Plex et al on something like a Raspberry Pi, but I want a bit more power. I want something that can push at least 2-3 4K streams.

I’m considering targeting 12th gen i5 with 16GB DDR4. Maybe 32GB if I can swing it. Is that overkill? I know you want 8th gen Intel for hardware transcoding of either x265/HEVC or AV1 (probably the former as the latter is newer).

There are other projects I’d like to run. I’d like to be able to host my own LLM/AI but I’m not trying to go performance overkill here. This would be my first server.

I am somewhat comfortable using the command line, and I would be remoting into the server via my Macs (I have a desktop and a MacBook). However, Linux can run Windows games via Proton and that’s something I’m interested in for some older titles that aren’t on Xbox (like Deus Ex, which came out in 2000 and should run on integrated graphics). I’m not inexperienced with Linux and have been using it off and on (mostly off unfortunately) for over 30 years. My go-to would be Ubuntu, but I’ve heard great things about KDE lately and I know the Internet is in love with Mint these days. Would the distro really matter all that much as long as I have the required services?

    • frongt@lemmy.zipEnglish
      1·
      2 个月前

      You can run stuff on a little GPU, IGPU, or even just a CPU. You’ll be limited by RAM/VRAM, and obviously it’ll be slower, but it’ll work.

      • Midnight Wolf@lemmy.worldEnglish
        1·
        2 个月前

        I have a server that I wanted to do this with, and cpu-only interference was several minutes for pretty simple tasks (all text, couple of sentences). I shelled out the cash for a 5060 Ti 16GB earlier this summer, and even with a pcie 3 x16 slot running at x4, I’m down to 5-30 second responses and rapid output speeds, even with large conversations. But the initial prompt/loading is still a couple of minutes by itself, so I leave the model in vram always.

  • glizzyguzzler@piefed.blahaj.zoneEnglish
    10·
    2 个月前

    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!

    • kayzeekayzee@lemmy.blahaj.zoneEnglish
      5·
      2 个月前

      I second debian!

      I initially had some trouble getting streamed games to use the integrated gpu while the server was headless. I’ve heard you can use a dummy plug into one of the hdmi ports, but I found that using “xrdp” as the remote desktop client works as well. I hope sharing this might save you the hassle evetually.

      • xavier666@lemmy.umucat.dayEnglish
        3·
        2 个月前

        I third Debian. I was a Ubuntu server fanboy for the longest time. I kept getting scared by “but it uses old packages” message about Debian. But it’s way overblown. Also, if you are using Docker for hosting your services, system packages don’t matter.

    • CerebralHawks@lemmy.dbzer0.comOPEnglish
      2·
      2 个月前

      Thanks. The more I see, the more I realise I should just leave the AI stuff out. AI and gaming were outlier possibilities anyway, though I’m way more interested in gaming than AI. Right there with you on AI ethics. I don’t like seeing them steal art and rip off artists, but they can be good at tracking down solutions on various forums. At least, they do it faster. But it sounds like running one might be outside my budget, so I’m more interested in running services, and light/classic gaming, since I have an Xbox (Series X).

      • horus_son_of_isis@lemmy.worldEnglish
        1·
        2 个月前

        I’m running a simple COMAI instance on my desktop that recommends linux commands and does limited chat. it was easy. It spins the processor up to 50% to give a response and takes a couple of seconds depending on complexity. I run the other stuff on a dedicated server, but with nginx proxy manager, I can send the COMAI instance to a page on my domain even though it is on a different computer on my LAN.

  • oyzmo@piefed.socialEnglish
    6·
    2 个月前

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

  • BartyDeCanter@piefed.socialEnglish
    6·
    2 个月前

    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.

  • B0rax@feddit.orgEnglish
    5·
    2 个月前

    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.

  • a1studmuffin@aussie.zoneEnglish
    5·
    2 个月前

    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.

    • CerebralHawks@lemmy.dbzer0.comOPEnglish
      1·
      2 个月前

      Yeah, I wasn’t targeting Intel specifically, that’s just what most/all the refurbished SFF PCs I’ve looked at have been running, and I know Intel is good for HEVC transcoding.

  • eleitl@lemmy.zipEnglish
    4·
    2 个月前

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

  • B0NK3RS@lazysoci.alEnglish
    4·
    2 个月前

    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

    • CerebralHawks@lemmy.dbzer0.comOPEnglish
      1·
      2 个月前

      What I have is a Mac mini (M2 Pro, 16GB RAM) and a MacBook Air (M2 base, 16GB RAM). The desktop (the mini) would be absolutely overkill, but I don’t want it doing a bunch of stuff when my wife and I are trying to actively use it as a server.

      The last Intel box I had shit the bed and I no longer have it. I had a 4th gen Xeon (1231-v3) with 16GB DDR4 and a 3GB Radeon 280. If I still had that tower, I’d start with that for sure.

  • pachrist@lemmy.worldEnglish
    4·
    2 个月前

    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.

  • JoeLong@lemmy.worldEnglish
    4·
    2 个月前

    Didn’t see it mentioned, but would recommend openmediavault. It runs on headless debian and is very stable.

    • FLAGSHIP@lemmy.zipEnglish
      1·
      2 个月前

      +1

      I’ve used OMV for a few years now. It’s relatively straight forward to just get working with plenty of options for more granular control. The documentation, in my experience, has also been quite good.

  • AllHailTheSheep@sh.itjust.worksEnglish
    3·
    2 个月前

    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”)

  • GreenKnight23@lemmy.worldEnglish
    3·
    2 个月前

    something that can push at least 2-3 4K streams.

    12th gen i5 with 16GB DDR4. Maybe 32GB if I can swing it.

    unless you’re transcoding each stream that should work, at least for Plex.

    Currently running 10th gen i5 with 32gb of DDR and can run one 4k transcode, five 1080 streams, and four live TV streams. though, the transcode is laggy but if you pause and let it buffer it should work.

    keep in mind I’m also running headless transmission, a VPN, tor, and like 8 other services on the same server.

    if all you’re streaming is 4k, you will hit a point of diminishing returns on network throughput. I’d recommend getting a switch with SFPs and connect your server up to a 2.5gbe or even a 10gbe (though overkill but the sky is the limit). as long as all your stream boxes are connected through that switch on their regular 1gbe they should be fine.

    personally I run a opnsense router/firewall and have everything passed through an unlocked Brocade ICX6610-48P-E. you can find them on eBay. FYI, these are commercial units and are VERY LOUD. basement or attic installation is a must if you don’t have a network closet. bonus, the ICX6610 allows you to turn on/off POE per port and even set the voltage. this means you can power everything from cameras, stream boxes, wireless APs, to even some POE lights. it’s also a managed switch, which means you can split your ports up on VLANs and other really neat things.

    • esperal@pawb.socialEnglish
      2·
      2 个月前

      Hi there ! If your i5 gen10 can only handle one 4k transcode, you might want to double check if your hardware acceleration parameters are correctly set ^^

      My home server runs on a small Intel N100, a 4k transcode eats up 30% of the CPU at most. I checked, it can transcode 3 streams without any lag, up to 4 transcodings if nothing else is eating the cpu. I’m not using Plex though, I got Jellyfin. I don’t know how Plex handles things.

      • GreenKnight23@lemmy.worldEnglish
        2·
        2 个月前

        my i5 doesn’t have a built in graphics core. it’s just a CPU. this means hardware acceleration for video encoding is intensive.

        like I said before too, there are other services that are running on the server as well. things like siem, aggregators, containerized desktops, etc.

        thanks for the info though! if I ever upgrade the CPU in the future I’ll definitely get one with a GPU for hardware acceleration.

  • mierdabird@lemmy.dbzer0.comEnglish
    3·
    2 个月前

    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

  • undrwater@lemmy.worldEnglish
    2·
    2 个月前

    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