Hmm, not sure why there hotze33, Magic is posting the same times for completion as I roughly, and I've had one of these 660's in two boxes now and the completion time was the same.
However, what comes to mind is the 470 being a Fermi which I'm inclined to believe are much better at CUDA than kepler are, considering the 550ti performance that I'm seeing (almost equal to the 660ti and much less powerful than the 470 you're running).
Considering that I've had the 660ti in an AMD box running 2.0 pcie 16x, and also in this Intel 3.0 box, and an additional identical card in the intel box... across the board performing identically, I'm inclined to believe that it rules out a general PC issue and is just the performance of fermi being superior to kelper.
Hmm, not sure why there
)
Hmm, not sure why there hotze33, Magic is posting the same times for completion as I roughly, and I've had one of these 660's in two boxes now and the completion time was the same.
However, what comes to mind is the 470 being a Fermi which I'm inclined to believe are much better at CUDA than kepler are, considering the 550ti performance that I'm seeing (almost equal to the 660ti and much less powerful than the 470 you're running).
Considering that I've had the 660ti in an AMD box running 2.0 pcie 16x, and also in this Intel 3.0 box, and an additional identical card in the intel box... across the board performing identically, I'm inclined to believe that it rules out a general PC issue and is just the performance of fermi being superior to kelper.
At least, that's my guess.
edit: Being curious, I did some brief searching and found this.
http://stackoverflow.com/questions/10765518/how-to-quantify-the-processing-tradeoffs-of-cuda-devices-for-c-kernels