I don’t think I am 100% sure about much of anything :-) As I said, I Googled the problem description and found multiple reports of slowdown without CUDA involvement on the first page of results. It is certainly possible that the use of CUDA toolchain exacerbates an existing issue. But as you can see from the log the “hang” happens when the host compiler is invoked on a .cu file, i.e. the CUDA toolchain is not in the picture at that stage:
I never use any IDE (or cmake, for that matter) to build CUDA projects. I am a dinosaur who is still familiar buiding with makefiles. No mysterious slowdowns there. The “building dependencies” is something the MSVC IDE does, and I am not aware that Microsoft has documented what exactly happens in that step. As a sanity check: Do you have any files that belong to this project that are located in huge directories? Any files that reside on (slow) network storage rather than fast local storage? A highly fragmented disk drive?
The only build “hangs” I have seen with the CUDA toolchain is when an optimization pass exhibits exponential growth in resource usage, e.g. a build chews through 8GB of memory and takes 20 minutes to compile one file. Those are basically bugs in the compiler, but I haven’t encountered any such issues in years. Since CUDA makes extensive use of function inlining and loop unrolling, sometimes a small-looking source file can balloon to several tens of thousands of instructions, causing longer compiles. But all those issues impact compilation proper, not “building dependencies”.