Nsight6.0 CUDA10.0.0: can't stop with breakpoint of kernel function

I’m using Nsight 6.0 Next-Gen CUDA Debugger with VS2017. But the debugger only stop when hitting breakpoint on host function.

My device is RTX2080Ti, and I have enable “Generate GPU Debug Information” option and set the “Code Generation” to “compute_75,sm_75”. I also add “-g” in the “Command Line” page.

What else should I do to make this work?

And NSight does not give me any error information. It seems like everything is fine…