Selfhosted

N...not quite...

Does this count ARMv6 256MB RAM running OpenMediaVault...hmm I have to fix my clock. LOL

I moved from a Drll R710 with dual docket Xeons to a rack mount desktop case with a single Ryzen R5 5600G. I doubled the performance and halved the power consumption in one go. I do miss having idrac though. I need a KVM over IP solution but haven't stomached the cost yet. For how often I need it it's not an issue.

7 websites, Jellyfin for 6 people, Nextcloud, CRM for work, email server for 3 domains, NAS, and probably some stuff I've forgotten on a $4 computer from a tiny thrift store in BFE Kansas. I'd love to upgrade, but I'm always just filled with joy whenever I think of that little guy just chugging along.

I run a local LLM on my gaming computer thats like a decade old now with an old 1070ti 8GB VRAM card. It does a good job running mistral small 22B at 3t/s which I think is pretty good. But any tech enthusiast into LLMs look at those numbers and probably wonder how I can stand such a slow token speed. I look at their multi card data center racks with 5x 4090s and wonder how the hell they can afford it.

Look for a processor for the same socket that supports more RAM and make sure the Motherboard can handle it - maybe you're lucky and it's not a limit of that architecture.

If that won't work, breakup your self-hosting needs into multiple machines and add another second hand or cheap machine to the pile.

I've worked in designing computer systems to handle tons of data and requests and often the only reasonable solution is to break up the load and throw more machines at it (for example, when serving millions of requests on a website, just put a load balancer in front of it that assigns user sessions and associated requests to multiple machines, so the load balancer pretty much just routes request by user session whilst the heavy processing stuff is done by multiple machines in such a way the you can just expand the whole thing by adding more machines).

In a self-hosting scenario I suspect you'll have a lot of margin for expansion by splitting services into multiple hosts and using stuff like network shared drives in the background for shared data, before you have to fully upgrade a host machine because you hit that architecture's maximum memory.

Granted, if a single service whose load can't be broken down so that you can run it as a cluster, needs more memory than you can put in any of your machines, then you're stuck having to get a new machine, but even then by splitting services you can get a machine with a newer architecture that can handle more memory but is still cheap (such as a cheap mini-PC) and just move that memory-heavy service to it whilst leaving CPU intensive services in the old but more powerful machine.

The beauty of self hosting is most of it doesn't actually require that much compute power. Thus, it's a perfect use for hardware that is otherwise considered absolutely shit. That hardware would otherwise go in the trash. But use it to self host, and in most cases it's idle most of the time so it doesn't use much power anyway.

I just upgraded to a Xeon E5 v4 processor.

I think the max RAM on it is about 1.5 TiB per processor or something.

It's not new, but it's not that old either. Still cost me a pretty penny.

Maybe a more reasonable question: Is there anyone here self-hosting on non-shit hardware? 😅

Just down load more ram capacity. It the button right under the down load more ram button.

Enterprise level hardware costs a lot, is noisy and needs a dedicated server room, old laptops cost nothing.

People in this thread have very interesting ideas of what "shit hardware" is

Maybe not shit, but exotic at that time, year 2012.
The first Raspberry Pi, model B 512 MB RAM, with an external 40 GB 3.5" HDD connected to USB 2.0.

It was running ARM Arch BTW.

Next, cheap, second hand mini desktop Asus Eee Box.
32 bit Intel Atom like N270, max. 1 GB RAM DDR2 I think.
Real metal under the plastic shell.
Could even run without active cooling (I broke a fan connector).

Somehow Jellyfin works ¯\_(ツ)_/¯

Not anymore. My main self-hosting server is an i7 5960x with 32GB of ECC RAM, RTX 4060, 1TB SATA SSD, and 6x6TB 7200RPM drives.

I did used to host some services on like a $5 or $10 a month VPS, and then eventually a $40 a month dedi, though.

7th gen intel, 96GB mismatched ram, 4 used 10TB HDD, one 12 with a broken sata connector that only works because it's sitting just right in a sled. A couple of 14's one M.2 and two sataSSD. It's running Unraid with 2 VM's (plex and Home Assistant), one of which has corrupted itself 3 times. A 1080 and a 2070.

I can get several streams off it at once, but not while it's running parity check and it can't handle 4k transcoding.

It's not horrible, but I couldn't do what I do now with less :)

I was for a while. Hosted a LOT of stuff on an i5-4690K overclocked to hell and back. It did its job great until I replaced it.

Now my servers don't lag anymore.

EDIT: CPU usage was almost always at max. I was just redlining that thing for ~3 years. Cooling was a beefy Noctua air cooler so it stayed at ~60 C. An absolute power house.

I had a old Acer SFF desktop machine (circa 2009) with an AMD Athlon II 435 X3 (equivalent to the Intel Core i3-560) with a 95W TDP, 4 GB of DDR2 RAM, and 2 1TB hard drives running in RAID 0 (both HDDs had over 30k hours by the time I put it in). The clunker consumed 50W at idle. I planned on running it into the ground so I could finally send it off to a computer recycler without guilt.

I thought it was nearing death anyways, since the power button only worked if the computer was flipped upside down. I have no idea why this was the case, the computer would keep running normally afterwards once turned right side up.

The thing would not die. I used it as a dummy machine to run one-off scripts I wrote, a seedbox that would seed new Linux ISOs as it was released (genuinely, it was RAID0 and I wouldn't have downloaded anything useful), a Tor Relay and at one point, a script to just endlessly download Linux ISOs overnight to measure bandwidth over the Chinanet backbone.

It was a terrible machine by 2023, but I found I used it the most because it was my playground for all the dumb things that I wouldn't subject my regular home production environments to. Finally recycled it last year, after 5 years of use, when it became apparent it wasn't going to die and far better USFF 1L Tiny PC machines (i5-6500T CPUs) were going on eBay for $60. The power usage and wasted heat of an ancient 95W TDP CPU just couldn't justify its continued operation.

4 gigs of RAM is enough to host many singular projects - your own backup server or VPN for instance. It's only if you want to do many things simultaneously that things get slow.

I'm sure a lot of people's self hosting journey started on junk hardware... "try it out", followed by "oh this is cool" followed by "omg I could do this, that and that" followed by dumping that hand-me-down garbage hardware you were using for something new and shiny specifically for the server.

My unRAID journey was this exactly. I now have a 12 hot/swap bay rack mounted case, with a Ryzan 9 multi core, ECC ram, but it started out with my 'old' PC with a few old/small HDDs

My home server runs on an old desktop PC, bought at a discounter. But as we have bought several identical ones, we have both parts to upgrade them (RAM!) as well as organ donors for everything else.

I started my self hosting journey on a Dell all-in-one PC with 4 GB RAM, 500 GB hard drive, and Intel Pentium, running Proxmox, Nextcloud, and I think Home Assistant. I upgraded it eventually, now I'm on a build with Ryzen 3600, 32 GB RAM, 2 TB SSD, and 4x4 TB HDD

Got all my docker containers on an i3-4130T. It's fine.