If programmer needs to assign work statically, does this mean programmer should break work into smallest pieces possible for balanced distribution of tasks?

I think the distribution balance of tasks could also depend on how heavy tasks are. As discussed in class, the number of blocks and threads per block is assigned by the programmer and should align with the capabilities of the GPU unit.

I would assume that rather than thinking about smallest pieces as possible, the goal would be to maximize the utilization of the GPU unit, as I think is demonstrated in following part of the lecture.