CPU efficient
nVector quantization by table lookup n16 memory accesses to find best matching vector.