A Large Intelligent Surface (LIS) is a recently proposed concept, especially suitable for high speed indoor communications and industrial internet of things (IoT) applications. Basing the LIS on smaller panels has clear advantages in terms of flexibility and mass production of its elements. In this paper we consider a panel-based LIS and we study the interplay of the panel size, the number of baseband outputs per square meter of deployed surface, the total activated surface area, the number of baseband outputs per panel, the terminal density and the ensuing minimum terminal rate. Our performance results show that it is desirable to employ smaller panels when the terminal density increases, but this means more outputs per m2, and higher overall LIS implementation complexity. It was observed that we can surpass such increase by working with higher fractions of the LIS area. Furthermore, we present an empirical equation stating the number of outputs per panel needed to ensure that all terminals are reasonably served. These results are useful for the LIS design in practical scenarios.