Unexpected behavior with more than 160 cuda kernels

Question

Basically when I run over 160 kernels at once the function seems to have sometimes executed twice on one kernel, when it should only be executing once. I\'m using a gtx 1050