Does the use of quantized data help make matrix computations faster?

Question

Given a quantized dataset $\mathcal{D}$ of $N$-dimensional data ; where only certain values are permitted for each vector component (for example $\mathcal{D} \subset \{1,2,3,4,5\}^N$), could this knowledge be used to speed-up matrix/vector operations which map any data point $x$ from $\mathcal{D}$ to $\mathbb{R}^N$ ($f : \{1,2,3,4,5\}^N \mapsto \mathbb{R}^N $) ?