Hello Cavi_Ing,
Tie points are actually sparse cloud points that you see in the Model view. They have at least two projections each, meaning that these are key-points that have been matched on two or more images (therefore "transferred" to tie-points).
When tie-point limit is used PhotoScan will leave only most reliable tie-points on each photo to fit the threshold, so it will results in lower number of sparse cloud points.
Using very high key-point and tie-point limits will result in longer processing time and may also affect the resulting alignment due to the fact that less reliable key-point could be matched.