SO, after my discussion in the other thread with End User, and further consideration on my automotive analogy (which, by the way, I still think was pretty valid), I had a brainwave.
Now, I don't know a whole lot about graphics (or, indeed, most computing) at a low level, so my thinking here may be unsound or based on fallacious assumptions, but it seems to me that we talk an awful lot about proper utilization of multi-core processors in a CPU context, yet it's never really discussed in terms of GPUs.
I assume, naturally, that this is because of the embarrassingly parallel nature of graphics work, since because of that it's relatively easy to utilize the entirety of even an enormous GPU like GK110. Still, I have to wonder; is there a relevant division between "single-threaded" performance (the performance of a single compute unit or shader core, influenced primarily by GPU core clock) and "multi-threaded" performance (the performance of the GPU as a whole, primarily influenced by core count?
Obviously nothing on a GPU runs "single-threaded", but perhaps a lighter load might be unable to utilize all of a big GPU's compute resources?
Microsoft claims that the Xbox One is competitive graphically with the Playstation 4 primarily because it runs its smaller GPU at a higher clock, discussions of that machine's eSRAM aside. Reading this, along with my own observations about my GTX Titan (where framerates are sometimes mediocre, but do not decrease when I raise settings, sometimes to absurd levels like 32xS AA),and further consideration of my analogy led me to wonder about the above question. Further increasing my curiosity are the facts that I've never seen anyone discuss this at all before, which leads me to believe that either A) I'm totally off the mark, B) everyone already knows this and I'm an imbecile, or C) I'm a genius, which is terribly unlikely. I'd love to hear some feedback from some folks with experience working with GPUs at a low level!