The memory consumption should be more or less the same as today. Then again, if there are plan regressions that starts to scan big tables, more pages may be brought into memory. If the case would be that is, that most of the data just sits on the disk and is never accessed as such.
Or more to the point: with knowledge of your workload, the question is not possible to answer. But if you cram in 64 GB of RAM into the server, most of the database will into memory entirely.