Scalable because this one can work in a vector length agnostic way. That is, you can write code that works with 128 bit vectors and also run on a machine with 512 bit vectors without recompilation.