When doing benchmarks we are usually interested in comparing the throughput of different solutions. However, when the code under investigation is not trivial, memory consumption may vary significantly, because using more memory to achieve better throughput is a typical design trade off. The talk discusses what memory metrics are more or less meaningful and what approaches can be used to track memory consumption alongside JMH benchmarks. A special focus will be given on dynamic effects on the interaction with the garbage collector and whether it is possible to get some stable benchmark results.