I have conducted the multithreading FFTW with pyFFTW in the python program, and I find that when the multithreading pyFFTW is running on 2D matrix, the responding number of cpu cores are occupied whose number is equal to the number of threads argumnet. However, the cpu utilization of each is so different, only one core's rate is high(~80%), and other cores' rate are low(20%~40%).
MY computing environment is ubuntu in VMWare.