Why cant the chip have knowledge of the capabilities of the GPU? Could be as simple as a GPU database with a gpu-z app running in the backend.