What has not been considered in the previous sections is computational speed. Since the τ-p method uses pre-computed tables, the actual computations are done long before the user decides to ask for a specific travel time. What the programs do on user demand is essentially table lookup, which works extremely fast, of course. TTBOX, on the other hand, was designed to avoid the use of pre-computed tables, because the need for such tables reduces flexibility. It therefore computes complete rays whenever asked for a travel time. Additionally, MatLab is an interpreted language and therefore slower than compiled FORTRAN.
TTBOX execution time scales linearily with the number of model layers covered by the ray and therefore with vertical sampling density and ray turning/reflection point depth. Execution time trades off with accuracy (please refer to the new section on Good & Bad Depth Sampling in the tutorial). For ray path plots and overviews of travel time curve sets, a crude discretization is often sufficient. More accurate times and therefore finer models are necessary for earthquake localization, especially on local and regional scales. These will take more time.
I do not give a table of times here, because the actual CPU time needed for a computation heavily depends on the used computer and its CPU load.