Qwen3-Coder 30B on a Surface Laptop Ultra: run it on RTX Spark with Ollama

My Surface Laptop Ultra decodes short code replies from qwen3-coder:30b at 90.9 tokens a second , twice the 44.7 that Microsoft's Lenovo Yoga 9n review measured for Qwen3.8-27B on the same chip. Both laptops use NVIDIA's RTX Spark N1X with 64 GB of LPDDR5X-9400. I repeated that post's two benchmarks with Ollama on Windows: no Docker, no vLLM, no speculative decoding. The long prompt tells a…