Qwen3-Coder local setup hits CPU ceiling

This looks less like a broken GPU and more like a model-size mismatch with a memory-bound runtime. Low GPU utilization here does not automatically mean the model is underpowered; it often means Ollama is juggling VRAM limits, context cache, and CPU offload.

• –Ollama's library puts `qwen3-coder:30b` at roughly 19GB and describes it as a 30B MoE model with 3.3B active parameters, so 16GB VRAM is already a squeeze. (https://ollama.com/library/qwen3-coder:30b)
• –Ollama's docs say larger context windows increase memory needs and recommend checking `ollama ps` for the CPU/GPU split; for coding tools, Cline recommends at least 32K context. (https://docs.ollama.com/context-length, https://docs.ollama.com/integrations/cline)
• –In practice, the fastest fix is usually not "more GPU usage" but a smaller model, lower context, or a more aggressive quantization for interactive coding.
• –For local VS Code workflows, Ollama + Cline is a legit stack, but 30B-class models are already at the edge of what a 16GB card can handle comfortably. (https://docs.ollama.com/integrations/vscode, https://qwenlm.github.io/blog/qwen3-coder/)