Two Qwen3 Models on One DGX Spark: The Residency Math for Local LLM Setup
www.devashish.me
The residency math, the gpu_memory_utilization trap, and what to verify first. Notes from my experiments with local LLMs.