Comment by bwasti

I've found the M1's 128bit instructions to be quite fast. My M1 MacBook Air can hit 90GFlops on a single core. My 2019 16" MacBook Pro is only 1.5 times faster at 135GFlops per core, despite double the vector width.

Benchmark here: https://github.com/bwasti/mac_benchmark

If anyone has the new 2020 MBP with AVX512 I'll update the benchmark to include avx512 instructions! I'm very curious about those numbers