The vast majority of the popular accounts are not run by the women on the profile. Most of them pay friends or agencies to manage the page for them, they simply show up to photo shoots every now and then and enjoy the easy money.

UI doesn’t come up until database migrations fully complete. Can take half an hour or more depending on how much content is indexed in your instance.

Your title should be “fuck subscriptions, except subscriptions from this site pulled from 1998” since everything in your guide relies on a paid debrid sub.

AMD has ROCm which tries to get close. I’ve been able to get some CUDA applications running on a 6700xt, although they are noticeably slower than running on a comparable NVidia card. Maybe we’ll see more projects adding native ROCm support now that AMD is trying to cater to the enterprise market.

Maybe don’t allow autonomous cars on public streets then? The tech is nowhere near ready for prime time.

Uhh… if your script is subbing to 24k remote communities, those will continue to grow from then on, unless you start purging communities at some point. After one user subscribes to a community, all new content gets indexed and stored on your instance. Pict-rs can cache images short term (and eventually clear them out), but Postgres will start growing very quickly and never slow down until it fills up disks.

The lawsuit does not involve Germany in the slightest

I’ve had access to 4 for several weeks and it’s not really much better. Maybe I’m just asking too much of it though.

Strongly disagree about gpt being excellent for code, it’s extremely confident about the wrong answer most of the time. I’ve found it to be mildly useful as a Stack Overflow alternative (for asking general questions and having it point me in some direction) but it’s code outputs are garbage.

These instructions won’t work in anyone’s unraid box, even if they compile compose from source. Not sure why people think posting random chatGPT’d instructions is remotely useful.

Article suggests you simply get blocked from watching additional videos. But there’s no info on how that works- is it account based? IP based? Can I wipe my YouTube cookies to bypass a block?

Lol. Guess it’s time to add the rest of my subbed channels to YT-DL and ditch their shitty ad-filled site entirely.

Apollo going away was the catalyst for me. I will never use Reddit’s garbage website or first-party app.

Plus Lemmy gave me an excuse to host another neat service and still waste the same time I did on Reddit.

I use Internetbs.net, sometimes Name.com if they have a particular TLD way cheaper.

I’m just letting mine do whatever it wants, got plenty of local storage. If/when I have storage issues I’ll add an s3 bucket, pretty easy to modify the entrypoint for pictrs to pass s3 connection info in the docker-compose deployment.

I’ve got a baremetal server with OVH running VMware, so it’s just a VM that I manage. I’m paying more for it than I’d like, but it’s running far more than just Lemmy. If I wind up ditching it in the future, it’s just a quick vMotion off to another machine + DNS updates.

Here’s a current output of my storage about a week into hosting the instance. It’s growing slower than I expected, and I do have plans to move volumes/pictrs up to an s3 bucket whenever I start running low on local storage.

[jon@lemmy lemmy]# du -sh volumes/*
2.5G    volumes/pictrs
2.2G    volumes/postgres

I would recommend locking down SSH on your Lemmy server, I have mine restricted to allow logins from VPN only. Otherwise you’ll get probed 24/7 with a public server.

Do you need access control? If not, a simple Apache/Nginx directory listing is nice and easy, just drop your files in your webroot and you’re set. h5ai is a nice addition if you go that route.

If you need access control (or at least some sort of obfuscated URLs), Nextcloud is a good option. Pretty easy to get up and running, and there’s a ton of plugins available.

From what I’ve seen and read, server to server traffic is less taxing on instances than client to server. So even if your instance is JUST you, it would be your instance talking to everything else so it would have some net benefit on the federation. But it would take a lot of users self-hosting solo instances for this to help in any noticeable way, I’d think.

There is certainly no downside to running a solo instance, if you’re even slightly interested I would say go for it!

I’m one of the other Lemmy.tf admins and I’ll share a bit. We’re currently on the docker-compose deployment from the repo, running on a VM with 4c/8gb ram/256gb disk. It’s on a baremetal VMware box at OVH with loads of resources to expand as needed.

I’m hoping we get enough users on here to force me into converting to a Helm chart and moving this to my Kubernetes cluster. Pod scaling would help address some of the issues larger instances are starting to run into, and it seems like a fun project.

As for Unraid, your best bet is to see if you can install docker-compose on it. This thread from 2020 suggests it should be possible, but the binary may not persist restarts. If you can’t use compose you would probably have to strip it apart and deploy one container at a time, and potentially work around the need for the Docker networks.

I may be interested in helping with an Unraid deployment guide if there’s heavy interest- I’m running it on my NAS at home and can tinker a bit. Feel free to DM me if you’ve got questions or need any assistance.

Edit: That Unraid forum post has a reply about using a bash alias to run docker-compose in Docker, this is the route I’d go rather than having to do jank stuff to make the binary persistent. Should be able to follow the normal docker-compose install from your root user once you have compose ready. Make sure to do your port forwarding or use Nginx Proxy Manager since SSL is mandatory to federate.

I’m interested, but I don’t know Rust and haven’t done frontend work in years. Might be able to do some work around scalability and contribute to a Kubernetes deployment guide (and/or Helm chart).