CUDA Streams problem

Hello to every one
I try to work with CUDA Streams using https://developer.nvidia.com/content/how-overlap-data-transfers-cuda-cc and receive interesting result.
I have 3 PCs with the next system configuration:
PC 1: GeForce GTX460 + Win7 x64 + CUDA Toolkit 5.0 + NVIDIA NSight Visual Studio Edition 2.2 + Visual Studio 2010
PC 2: GeForce GTX660 + Win8 x64 + CUDA Toolkit 5.0 + NVIDIA NSight Visual Studio Edition 2.2 + Visual Studio 2010
PC 3: GeForce GTX470 + Win8 x64 + CUDA Toolkit 5.0 + NVIDIA NSight was not installed + Visual Studio 2010
When I start the program in profiler using PC3, I see overlapping:

But the same program using PC1 and PC2 doesn’t show overlapping in profiler (fig.2) and the time is the same as version without streams.

Currently I try to uninstall NVIDIA NSight and reinstall CUDA Toolkit to the last version on the site (January, 2013). But I’m not seeing overlapping again.
Any ideas what’s wrong?