Just wrote a post about it Some will say this is good for mixing and matching hardware, ignore them, as this will be way slower than Unified Memory Besides that, great option for single GPU inference

→ View original post on X — @theahmadosman