Box Collision vs Static Mesh simple collision (the static mesh is a box)

GPU collision? I thought all collision was CPU based.
What I do is lerp the scale of the collision box between 1 (default) and the target scale. Then I pass this variable to the niagara so that the GPU particle travelled distance in the X axis follows the collision box.

Still, the question is what is more performant, a box collision object or a static mesh object with simple collision (not complex collision).