Gpu

Every other self-hosted app on this site replaces a subscription with something you run yourself. AI is a little different, the frontier cloud models are genuinely ahead of anything you can run on homelab-grade hardware, and pretending otherwise sets you up for disappointment. What self-hosting AI actually buys you is privacy for anything you don’t want leaving your network, zero per-token cost for high-volume low-stakes tasks, and a model that keeps working when your internet doesn’t. Ollama and Open WebUI are the two pieces most homelabbers land on to get there, and they’re worth understanding separately before you stand up either one. ...

Buying the right GPU for your homelab is one decision. Getting Proxmox to actually hand that GPU to a VM or container is a separate one, and it’s where a lot of people get stuck, because Proxmox actually supports two completely different ways to do it, and picking the wrong one for the job means either a fight with IOMMU groups you didn’t need to have, or a container that can’t do what you needed a full GPU for. Here’s what each path actually requires, and how to pick between them. ...

“Do I need a GPU for this” comes up constantly in homelab planning, usually attached to one of two very different goals: hardware-accelerated transcoding for Plex or Jellyfin, or running local AI workloads like Ollama or Stable Diffusion. Those two goals point at completely different hardware, and buying for one when you actually need the other is one of the more common expensive mistakes in this hobby. Here’s how to figure out which category you’re actually in before you spend anything. ...