I don't think it's a library issue; I'm running nvvp on a Windows machine using CUDA 4.2 and I'm not having any trouble. It could be that Euler's library for that isn't quite up to date. I don't think it'll make a big difference whether you use the 4.1 or 4.2 version of nvvp. That said, you can always export your executable and run the profiler on a different machine. The CUDA toolkit for Windows comes with an exceptionally easy to use profiling tool where you can choose a project or just an executable and command line arguments. Hope that help!