All measurements are taken directly from each card, since monitoring system-level consumption is not exact enough. After all, different graphics loads affect processor utilization, causing platform power to fluctuate.
Of course, we are especially interested in a direct comparison between the GeForce GTX 660 Ti, GeForce GTX 670, and the new GeForce GTX 760, since all three models compete against each other, albeit indirectly.
At the outset, all cards behave similarly.
Stepping up to a gaming workload reveals greater differences, although the new GeForce GTX 760 once more falls between the GeForce GTX 660 Ti and the GeForce GTX 670. That makes sense, considering its clock rate. Factory-overclocked versions of the 760 come much closer to the GeForce GTX 670 both in power consumption and performance.
In compute-intensive applications, the GeForce GTX 760's power consumption tends closer to the 670 than the 660 Ti, even at stock clocks. Considering the new card’s mixed performance in our CUDA and OpenCL testing, that’s not really justified. The GeForce GTX 760 never hits its thermal or power targets, so our 80-degree scenario only becomes a limiting factor for the higher-end cards.
As long as the GPUs don't exceed their rated thermal or power limits, they hit power peaks beyond Nvidia's TDP. In practice, this happens very rarely, and only briefly at that. Still, don’t forget to take those situations into account and pick your power supply accordingly.