I’ve yet to see a fair comparison/assessment on M1 Ultra’s GPU performance in terms of ML potential. The same with 3D rendering outside of Blender 3.1 speeds which is not optimal (I also wonder if industrial rendering engines such as Renderman will support ARM in the future). As more and more softwares optimize for ARM, I see a great deal of potential for these machines above and beyond what is possible right now within limited ecosystem. I am 100% convinced Apple struck gold with this.