Zimaboards do not need RAM. They include it (do not confuse it with the Zimablade).

I just use the default hardware with the Zimaboards. Zimaboard 832s for the masters and nodes, and I recently added a couple of Zimaboard 2s for larger nodes as those have 16gb.

I like the boards, they perform well and the intel cpus in them have internal GPUs so for light transcoding or GPU loads they can be leveraged without need for GPU cards. After trying several other x86 based options, Zimaboards are my preferred option.

I used Zimaboards. They are affordable and all you need if you couple the cluster with a TrueNAS Scale box and setup Democratic CSI.

Why not both?

Hahahahahahahahahahahahaahahhahahha…

deep breath

Hahahhahahahhahahhahahhaha…

Fuck off you twat. Your chatbots are stupid.

Guillotines and a lineup of billionaires in straight jackets

Complying with this shit is nuts.

Is there an Arch fork that is not implementing this shit or do I have to go non systemd now? Because this BS is not going on any of my machines.

Fuck AI.

Personally, I think this first response nails it.

https://lists.ubuntu.com/archives/ubuntu-devel/2026-March/043515.html

Linux is not sold. So you either need to force users to install this on their systems, or go eat rocks.

Enterprise distros on the other hand… Need to do this.

Kube makes it easy to have a lot, as a lot of things you need to deploy on every node just deploy on every node. As odd as it sounds, the number of containers provides redundancy that makes the hobby easy. If a Zimaboard dies or messes up, I just nuke it, and I don’t care whats on it.

Kids should NOT be on social media.

Uh… Probably somewhere around 150?

Was super interested right up until the AI recommendations bit.

Lol, lost the internet because they have less DCs?

Meanwhile most of America is still on dialup or early 2000s broadband speeds.

This is a fucking shallow definition of winning.

I have run Pihole on 2 physical Pi 4s (DietPi OS) with config sync for 3 years now. Core to the house. Very reliable.

Oh sorry, just realized we are talking app servers.

Yeah, Google apps, and linux hosted apps. Havent had a company that ran windows or MS anything in 14 years.

Forgejo and self hosted action workers.

Mac, actually. Its a different kind of bad. At least I can use many of the same cli tools.

Its also trained on stolen data, artists work without their permission. AI training, even for the offline models, uses massive amounts of electricity and water and is currently accelerating climate around the world as well as unaffordability as demand for water and electricity cause prices to skyrocket. At the same time its accellerating the unaffordability of personal computing, including phones, and threatening to remove open PC hardware platforms by removing direct access to affordable DIY hardware.

On the other side of this, continued use and justification of LLMs existence is enabling the founding of mass surveillance and control systems that will be the foundation for totaltarian states, while at the same time enabling the rich to manipulate and control truth. And because of randomized token tie breaking, anything that comes out of it is only partially correct even when its one of the 30% of the times the reply is partially useful.

And - on top of all of that, you are nerfing your own skills and brainpower everytime you use it, in addition to having it do something for you that you could be learning yourself, which would have increased your existing skills while teaching you a new one.

AI is a horrible technology, doesn’t matter where you run it.

Lets fucking hope not.