I don’t think there is any cuda debugger. You can compile using device emulation where it will create code which runs on your computer (not on the actual cuda card), simulating running on a cuda card (but of course much slower). Use the -deviceemu nvcc flag to enable this. Then you can use any normal c debugger to step through the program.
The only other way of debugging code is to copy back the memory from the cuda card and look at the values after the kernel completes.
You should be able to open the *.cu files in any text editor. It is just c code.
The CUDA 2.1 release now includes a debugger as a separate download, but it is only available for 32-bit Linux systems. 64-bit Linux and Windows debuggers are not available yet.