I know that some of them are binned, like the ones used in GTX 580 classified ultra, but I'm not sure how often binning is done.
 
I think what they do for the SC and FTW type ones is that they make the card, and see if it can meet the OC specs, passing all the tests, if it does, it's a SC version, if it doesn't but does pass the tests at normal clocks, it becomes a normal chip. if it fails at normal clocks, it's a reject I guess.
 
however some aren't checked beyond standard clocks and as such one might get a normal clocked card that can OC like a champ, or one might get a normal clocked card that didn't quite make the cut to be a OC version.
 
Also I think some are binned by Nvidia before being shipped to manufactures,  as I recall some chips where binned for the 500 series for use in GTX 590s, while others were not quite up to the challenge.
 
like I said before, even though there may be a few that are rejected from the highest levels of testing (of the OC versions or whatever) not all are, so you might get a card normal card that has a FTW card quality chip in it, because not all are tested beyond the stock clocks.

I would be interested in learning how they do this.  I would suspect it like what bigbrag says or possibly some sort of QC specs and tests done to the waffer itself.  Who knows but I would like to see an article on it so I really knew.

The reason I ask is because there is no way they just take each card, pop it in a system and test it. Plus, yields from batches can vary. I remember reading that generally, dies that are near the center of the wafer during manufacturing tend to be the golden chips whereas when you get closer to the edge of the wafer, quality begins to suffer. I know component quality also matters in the whole scheme of the card, but the gpu dies themselves are what I'm interested in.
 
I assume they have to have some way of quickly and easily binning most dies after manufacturing. Also, when they disable clusters and such for cards that don't meet the high-end criteria (think a 680 chip that doesn't quite cut it becomes a 670, which has one cluster disabled) how are they disabled? I know a good while back Nvidia started laser cutting traces so they couldn't be reactivated (I'm talking a GOOD while back, like after the FX 5900's)

I'd be really cool to get one of those "Insider" YouTube videos.
 
I'm pretty sure they have some sort of machine that has terminal points that make a solderless connection with the needed points on the chip, push some volts through it, and it can read what it's capable of.  At least, that's what I imagine would happen.
 
Again, it'd be really cool to have a video 

Check out the link I posted above to look at something like they use for testing., we do this kind of thing hundreds of time a day at work, it really isn't all that amazing, lol. Stick IC in socket, power up, run 30 second test, take it out and put it in the dry box lol.
Integrated circuits are never put into the system that they are intended to be used in when coming from the manufacturer, EVGA or other third parties may do this on select chips once it has been assembled into a finished product, but everything is placed into a socket like the one that I linked above and is tested in that matter. Very few IC's are binned from the main manufacturer, because as long as it works as advertised, that is all that matters, quality often times takes a back seat to production and the amount they ship. Most companies worry about shipping and making their profits, and if it happens to come back for whatever reason, they will cross that bridge when it gets there.
I have a programmer / tester at my house for DIP (dual inline package) IC's, specifically made by Microchip and their series of PIC 8 and 16 bit microcontrollers, particularly the PIC 12, PIC 16, and PIC 18 platforms. However more commonly we are using ICSP or ISSP, In Circuit Serial Programming, which uses a debug interfaces that can test and program the chips once they have been soldered onto the pcb, because more often that not, once you have soldered the part down through reflow, and it has to be heated again to be removed, often times with a hot air gun, all of the heat placed on the part can cause problems. All integrated circuits are on a JEDEC standard which specifies the maximum amount of heat and time at that temperature that an IC is rated for prior to failure occuring. If you would like me to post some pictures of my programmers and testers and such, plus my entirely too large collection of ICs I will do so upon request.

EVGA_JacobF

Today, we do not do binning on the just the GPU. There are a lot of components on a graphics card that have tolerances, so it is not good to just rely on GPU only.

We test the entire board, 100% of them. If pass, its gets the SC badge :)

 
Solid answer!  Thanks! 

But there is a binning process on just the GPU die that occurs on the factory level, this is way early in the process though, and done more on NVIDIA side.