Dynamic programming is often useful to speed up FPT algorithms. It uses, however, often a lot of memory. We show that the high space complexity is unavoidable by tight lower bounds on several problems.