Ollama hits AMD R9700 compatibility wall

This looks less like user error and more like the usual AMD local-LLM tax: the GPU stack works in one app, then breaks in another because the backend support is still uneven.

• –LM Studio working with the same card and ROCm install is a strong signal that the hardware and base drivers are at least partially functional
• –Recent Ollama GitHub issues describe the same R9700/gfx1201 discovery timeout and CPU fallback on other Linux setups, including Docker
• –The real friction here is app-level ROCm support, not just whether AMD advertises ROCm compatibility for the card
• –For AI developers building local inference rigs, new AMD GPUs still require checking support tool by tool, not assuming the whole stack is ready on day one