I can get 2.5 spark for the price of the M5, will have better throughput and access to bigger models (more vram when running tensor parallel)