Just to add a little comment about the internal workings of it:
The code divides the image height and width by the "tile size" input, and then rounds each to the next highest integer, which are then again used to divide the image into tiles.
This is done to have a consistent tile size, and not end up with a "leftover" tile that is something like 60 pixels wide and causing artifacts.
So if you are testing this feature, don't expect different results for small changes, like tile size 1000 and 1050