Program runs with cuda memcheck but not with cud a gdb?

I have a CUDA C program that compiles fine and runs perfect when I use the cuda-memcheck tool. However when I try to run it without cuda-memcheck or with cuda-gdb I get Lane Illegal Address?

What kinds of problems can generate these errors?

Thanks

If I enable blocking mode (–blocking) on cuda-memcheck the error is caught.

I guess this is because in cuda-gdb when cuda-memcheck is used blocking is turned on by default?

I am still interested in what kinds of problems generate these errors. I am not 100% about blocking vs. non-blocking.

Thanks again