thanks a lot, yea it makes sense.
Edit: it seems weird when i think that on the blog it takes 490 instructions whereas in my case it takes 360. That is roughly 26% less instructions required than on blog.
Instructions per cycle are more or less same on both results, telling me that my M1 Air is somehow using less instructions to run the same code (or is faster)
I obtain nearly identical results to yours on my M1 MBA (16GB). Maybe his machine was swapping while running the benchmark?
Author has posted an update matching these figures—-they were using an earlier build of their profiler