Originally Posted by fateswarm
To those "feeling" 2 GPUs are more horsepower (de facto), there's a simple information bit you could use to shed that myth: GPUs are by design multithreaded. The whole thing that makes them needed compared to a CPU is that instead of getting a whole bunch of info and calculating it one by one, they get a bunch of very closely related info and just calculate it simultaneously.
Since there are tradeoffs, they are worse than CPUs for complex calculations of a singular nature or an evolving nature, but they are better for calculations of a simpler nature that are done in parallel with other calculations that are similar and share a large part of the assets. e.g. CPU: better for 100 calculations in a row that use 100 completely different algorithms (especially when they are of a complex nature). GPUs: Better for 100 calculations that use a similar algorithm (especially if it's a simplistic one).
As a result, a lot of "Better" GPUs are nothing but a "Worse" GPU that was just expanded to more processing units or let's just call them "cores" for simplicity. I'm surprised at how easy NVIDIA and AMD have it when they produce a more "advanced" GPU (of a similar generation). One can easily see that they often just increase chip size and duplicate the same things a few more times. More expensive? Yes, but simply because of spending more Wafer space, not so much for design/R&D.
Basically you can easily see how GPUs are largely designed if you do some GLSL programming. Most of what you're doing is "Do this a bunch of times in almost the same way and then do that a bunch of times almost the same way" and so on. They are simpler but highly parallel beasts. Though of course, a lot of work must be needed to optimize that parallelism.
And definitely, it's better to have that parallelism on one chip rather than two. So you do want one chip over an SLI setup that provides the same assets. It's not an accident that SLI/CF are often deficient compared to the same power on a single chip.