from what i can understand, fermi has shaders that run at 2x clock speed. so cuda core per cuda core, yes, the fermi is faster. however the kepler's "wide" philosophy is supposed to provide significantly decreased power usage while maintaining gaming performance (at the expense of compute performance)