I've got a case where subsequent (adjoint) NFFTs on GPU are giving different results. I don't have a MWE yet but just to prove it:

Is this something which one could expect to happen in any scenario? Perhaps I'm doing something wrong? Any hints that are possible from the little info I provded might help me track down the MWE. In any case trying to get something...