A home lab enthusiast deployed the 35B‑parameter Qwen3.6‑A3B model on an RTX 3080 Ti by offloading workloads to the CPU and tuning system parameters, sustaining over 20 tokens per second. The system ...