I am new in cuda and learning it on a MacBook retina with gf650m. Via profiling have I seen that I have a very varying execution time of my kernels. I assume that is because of the energy saving mechanism of the gpu. As example: If I execute a little kernel (a runtime with only a few ms), the execution time is relative high. If I execute the kernel several times in a row, the execution time goes down. If I move a window or start a 3D Application, the execution time goes also down.
Is there a method to turn off of the energy saving mechanism or is there an other problem?
and ps: sorry for my bad English :(