AnalysisAI ModelsSeptember 19, 2026

Home server built from old PC runs Qwen3.8 27B at ~30 tokens/sec

A Reddit user repurposed an old PC with a GPU upgrade to run Qwen3.8 27B locally at roughly 30 tokens per second. The setup was built for a single project, giving the model access to project files to search for bugs.

1 source

More stories today

Open the live feed
Home server built from old PC runs Qwen3.8 27B at ~30 tokens/sec