Exact block size is not necessary, big safe choice is fine too. But I suspect it's trivial to detect it with benchmarks, as writing even a byte over a block size would require overwriting two physical blocks instead of one, which is much slower.