At most my Strix Halo uses $0.48 a day

This is something that never gets mentioned when people complain that it's slow and new users are told to avoid them. This 48 cent figure is worst case scenario, running multiple models/compiling hitting CPU, GPU, and NPU at the same time for 24 hours a day. I can handle only 50tps on Q8_XL Qwen 3.6 35B when it's silent, sipping power, and is the size of a small router. I know your Nvidia card is significantly faster, but if you even consider using more than just raw GPU memory speed/compute or you are conc