Thanks for reporting this. As a workaround, please use j0(x) instead of jn(0,x) as njuffa suggested earlier. Note that the same problem applies to jn(1,x); in CUDA 5.5 and earlier, you will need to use j1(x) instead.
We will correct these to be properly equivalent in a future release.
EDIT: Sorry for the duplication; I hadn’t seen the second page of followups before posting my reply.