> It's faster yes

...for (non-SIMD) integer operations. For floating point and memory speed, it seems to be the other way round.