Ok. So my second point is incorrect - the GPU does not poll the CPU, the CPU sends requests to the GPU to do some work. Easy enough.
Still, a 680 GTX is going to process requests much faster than the non-Ti 750. I wonder if there would be as big of a performance hit. Using made-up numbers, let's say the non-Ti 750 spit out 45 fps with an i7/i5 setup up...would it produce the same 45 with an i3, or would it drop to say 42 fps (a much smaller hit)? Obviously, the i3 CPU wouldn't be able to send as many instructions to the GPU, but then again the GPU wouldn't be able to process as many commands to begin with.
So I'm doing some investigation regarding this and I'm having trouble defining a point where the CPU is still bottlenecking. If the CPU is without a shadow of a doubt at 100%, it's bottlenecking. But I've found even at 80%, it's still bottlenecking. I don't know if this is just a scheduling quirk or if the sampling rate is too low.
This is kind of a cop out, but the easiest definition of bottlenecking is when a CPU can no longer process efficiently causing other system components to be underused. Of course, the harder part is determining when
this occurs. I think it would be reasonable that any usage above 90% would cause some type of bottlenecking. Between 80 - 90%...? Hmmm, I would think there's enough breathing room that it wouldn't occur (or it would be very minimal), but I'm not a circuit-level expert and if you're seeing it at 80% that is interesting.