← Back to Build Journal

Today, the Beelink SER9 became the first working node of the AI Lab. Here's everything that happened, in order.

Hardware

MachineBeelink SER9
CPUAMD Ryzen AI 9 HX 370
RAM32 GB LPDDR5X
GPURadeon 890M integrated graphics
NPU~50 TOPS AI NPU
Storage1 TB SSD
Browser download popup showing OllamaSetup.exe downloaded
Ollama installer downloaded

The least dramatic part of the whole day — and after everything else that happened, that's a compliment.

Setup completed

Ollama setup wizard, version 0.31.1, ready to install
Ollama installation in progress

"Click Install to continue" — the last easy decision for the next twenty minutes.

📷 Photo coming soon
AMD ROCm/HIP library observed
Command prompt output of ollama --version showing 0.31.1
Ollama version verified

The moment you find out the install actually stuck, instead of just looking like it did.

Command prompt running ollama run qwen2.5-coder:14b
Qwen2.5-Coder downloaded

Typed the command, held our breath. First time asking this box to think for itself.

Terminal showing Qwen2.5-Coder's first response, with disk space after download in the background
First local Qwen response

It's alive, and it introduced itself before we even asked. First real proof the whole stack — model, Ollama, hardware — was actually talking to itself, not just installed side by side.

Storage measurements

Before Ollama867 GB free / 930 GB total
After Ollama862 GB free
After Qwen2.5-Coder:14B853 GB free
After Open WebUI + deps847 GB free

Approximate usage

Disk usage table: before Ollama 867 GB, after Ollama installed 862 GB, after Qwen 2.5 Coder 14B 853 GB, total AI stack so far 14 GB
The receipts, straight from the disk

14 GB for an entire local AI stack. Cheaper than installing a AAA game.

📷 Photo coming soon
Tailscale device ai-lab-b connected
📷 Photo coming soon
Open WebUI detected qwen2.5-coder:14b
📷 Photo coming soon
Open WebUI chat interface

Quote of the day

"We're on a mission from Allah."

(A joking Blues Brothers-style moment — not to be taken any more seriously than the movie was.)

Lessons learned

No separate AMD ROCm install was needed — Ollama brought its own ROCm/HIP libraries for the Ryzen AI platform. See Engineering Notebook, Note 001 for detail.