I seems that the cuda openMP example is not getting compile under windows with openMP support!
This lead to the example to only launch one thread… and not the a number equal to the number of GPU…
I appears on visual studio 2008 SP1 (windows 7 and windows XP)
In order to solve it one as to go to
Project Properties > Configuration Properties > CUDA Runtime API > Host > Extra C++ Options
to add /openmp
I hope this will help some people.