An unordered multimap works in a manner similar to a multimap, except that the internal behavior of these containers differs. A multimap makes use of red-black trees while an unordered multimap makes use of hash tables. The time complexity of multimap operations is O( log N) while that of unordered multimap operations is O(1); hence, an unordered multimap tends to be faster than a multimap.
The values stored in an unordered multimap are not organized in any particular fashion, unlike in multimaps where values are sorted by keys. If performance is the criteria, then an unordered multimap is a good bet; however, if iterating the values in a sorted fashion is a requirement, then multimap is a good choice.