Seems like the classic trade-off of compute time x memory.