1023 gflops on the M1 mini using the kernel in the blog post, although if I let it run some more it discovers an 8-way unroll that's a bit quicker (1073).
I just upgraded from a 2017 12" macbook to an M1 air. It's day and night in terms of performance across the board. I love the fanless design and the M1 air runs cooler than the 12". Also the keyboard is much better.
Another ridiculous comparison - 70.2 gflops on a iPhone SE 2016 (actually it hit 109 at one point, but that seems to be a fluke, whereas ~70 was hit fairly reliably).
