Built My Own AI Server at Home — Here’s the Setup and Why It Rocks

4/26/2025•Qorvinus
AIDevOpsLLMsVPSBackend

Why I Built This Server

Between dabbling with local LLMs, playing with AI image generation, and wanting more control over my own data, I decided it was time to stop renting compute and build something powerful at home. I already had a Ubiquiti Dream Machine Pro handling my network, so it felt like a natural next step to integrate a local AI powerhouse.

The Build: Specs and Setup

  • I went a bit overkill, but no regrets:
  • CPU: AMD Ryzen 9 9950X3D — 16 cores and beefy cache for parallel workloads
  • Motherboard: MSI MAG X870E Tomahawk WiFi — rock solid with future-proofing
  • RAM: 64GB DDR5-6000 — might double it to 128GB later
  • GPU: NVIDIA RTX 5090 — absurd amounts of VRAM for local LLMs and image models
  • Storage: 2TB NVMe SSD — fast and enough room for models
  • Networking: 5GbE LAN, WiFi 7, Bluetooth 5.4
  • OS: Windows 11 Pro — for now using WSL but might dual-boot

Network and Access

I assigned it a static IP and set up SSH access for remote management. Thanks to the Dream Machine Pro, routing and reverse proxying within the LAN was super smooth. It now behaves like any other always-on server, but tailored for AI workloads.

What It's Running

Right now, I’ve got:

  • ComfyUI — for stable diffusion workflows and image generation
  • OpenWebUI — as a slick front-end for interacting with models
  • Ollama — for running LLMs locally with a simple CLI and API

Everything is containerized where it makes sense, but I kept some native installs for performance reasons.

Challenges and Fun Bits

  • WiFi 7 support is cool, but I still wired it up over 5GbE because latency matters.
  • Power draw is non-trivial, especially with the GPU running long jobs — something to monitor.
  • Windows is surprisingly fine, but managing drivers and CLI tools takes more effort than on Linux.

Tools and Tech Stack

  • Hardware: AMD, NVIDIA, MSI
  • Networking: Ubiquiti Dream Machine Pro
  • AI Tools: ComfyUI, OpenWebUI, Ollama
  • OS/Access: Windows 11 Pro, SSH, static IP, reverse proxy

Things I’d Improve Next Time

  • Might add another 64GB of RAM for larger model runs.
  • Considering adding an external NAS to separate data and model storage.
  • Eventually want to automate some of the container and model updates via scripts or n8n.