The same collision problem, three ways of deciding which pairs are even worth testing.
What to notice. Push the count up with brute force selected and watch the pair-test number climb as the square of it. Switch to the grid: same objects, same collisions found, a fraction of the tests. Then set the cell size far too small or far too large and watch the saving evaporate — cell size wants to be about one object wide. The quadtree costs more to build but doesn't care how clumped things are.
← all demos