unordered_map not working with CUDA?

[Oblynx]

There is a problem with using std::unordered_map (instead of std::map) in TaskServiceStatic, but only in the CUDA example. When the same code is included into normal C++ it works.