Comment by hamiltont
6 hours ago
Once you hit the memory you need, generation speed is mainly set by bandwidth, and every Ultra from M1 thru M3 has ~800 GB/s. IMO best ROI for most people is 'cheapest used Ultra with enough RAM'
I setup an eBay alert and picked up a used M2 Ultra that has delivered good ROI (at least, far better than 15k for comparable-for-my-use-case performance)
I think M1 through M3 were compute bottlenecked in prompt processing (hence the very large gap between M3 and M5, in this page's benchmarks, that's not explained by memory bandwidth alone).
For generation speed in isolation, yes.
The M5 generation added tensor instructions to the GPU cores.
This one is 2x m5 max, so ~1.2TB/sec.