A few months ago, I mapped out a "5 Phase Roadmap" to build a high-performance Home Lab. The dream was big: dedicated enterprise servers, 10Gbps networking, and a massive GPU cluster. But plans don't run Docker containers. Waiting for the "perfect" hardware was stopping me from starting.
My Home Lab: From Zero to Running Containers
A few months ago, I mapped out a "5-Phase Roadmap" to build a high-performance Home Lab. The dream was big: dedicated enterprise servers, 10Gbps networking, and a massive GPU cluster.
But plans don't run Docker containers. Waiting for the "perfect" hardware was stopping me from starting.
So, I’ve torn up the complex roadmap and replaced it with a pragmatic, efficient, and surprisingly powerful Phase 1. We aren't building a data centre; we are building a Hybrid Cluster.
The Strategy: Brain vs. Muscle
Instead of running three mediocre servers or one massive power-hungry one, I’m building a 2-Node architecture optimised for power and silence.
Node 1 (The Brain): A highly efficient, always-on server for smart home control and web traffic.
Node 2 (The Muscle): A recycled gaming beast that only wakes up when I need heavy lifting (AI or 4K transcoding).
The Hardware: The Build List
I’ve pulled the trigger on the foundation: a 15U Open Frame Rack. It’s simple, cheap (£90), and provides the verticality I need. Here is what is going inside:
Node 1: The "Always On" Controller
Role: The 24/7 Operations Centre. Runs Home Assistant, Traefik, and the Vyzo discord bots.
The Tech: I’m building a custom node using the Intel N100 chip (ASRock N100M motherboard).
The Case: A sleek 2U Logic Case (SC-2340).
Why: This machine is silent and sips just 6-10 watts of power. It keeps the house running without spinning the electric meter.
Node 2: The "Franken-Server"
Role: The Heavy Lifter. Runs Jellyfin (Media) and Ollama (AI).
The Tech: I’m recycling my old i7 gaming PC internals, including a GTX 1070.
The Case: A massive 4U Logic Case (SC-4450). We needed the 4U height to fit the GPU vertically.
The Magic: Using Proxmox LXC Containers, I can share the single GPU between the media server and the AI engine simultaneously.
The Spine:
Networking: A TP-Link 16-Port Gigabit Switch. No loud PoE fans yet—just reliable, silent copper.
Power Strategy: Node 2 will be configured with Wake-on-LAN. Node 1 (via Home Assistant) can wake up the "Muscle" node only when I need to process a video or chat with the LLM, then shut it down automatically at night.
Why This Works
This setup gives me the best of both worlds. I get the low-latency, "always-up" reliability of a dedicated server for my smart home, but I still have the raw CUDA power of the GTX 1070 for my Vyzo development and media streaming.
Most importantly, the entire build comes in at roughly £500, repurposing hardware I already own to keep e-waste down and value up.
The parts are ordered. The rack is on its way. It’s time to start building.