Here's what I think (roughly) typically happens ... at least amongst the companies that actually MAKE some of their cards, like GB, MSI, Asus, and Galaxy ... as opposed to just being a re-branding company like evga or xfx.
Say you're Gigabyte, and you receive 10,000 GF114 cores, and your marketing people tell you that you should make 2,000 of them into the SoC models, based on sales projections and such.
Now, I believe GB probably has some testing tool(s) that can determine, within a reasonable margin, which chips are likely to be able to hit some number of MHz, without actually building the cards. Seems to me that they'd have to have some way of doing this.
So, I believe what they'd do is randomly test the chips, and the first 2,000 they find that pass the tests, such that they appear likely to support the SoC clocks ... those would go onto the SoC PCB's (with their upgraded components), and become the required 2000 SoC 1GHz cards.
Now, speaking to this specific case: since the SoC is really quite a high OC'd card relative to reference, and GB has to decide which chips they're going to use prior to building the cards since it's an upgraded component/non-reference PCB, there's no doubt that once these cards are actually put together, the initial estimates of which chips will actually do 1GHz ... will turn out to be 'wrong' in some cases.
Ergo, a new SKU comes into existence, that of a 2nd SoC model that only runs at 950MHz.
Now, certainly this is a form of binning, but it's not the binning that people around OCN are talking about.
The binning process that we would all be hoping for is that they'd test all 10,000 chips, figure out the 'Top 2000' in terms of the chips able to reach the very highest clocks at the lowest possible voltages, and then turn those into their top o' the line SoC cards.
Personally, I doubt that a binning process like the one that we would hope for happens very often, if at all. You're talking about a bunch of extra expense/work, and probably much more sophisticated testing equipment ... w/o a clear-cut benefit from doing so.
I mean, I suppose you'd have the 'geek' crowd amongst your employees that would be like "Yeah, binning the best of the best for the SoC model makes sense because the enthusiast community is going to 'notice' that the SoC is not only the fastest OOB, but that they also are capable of the highest overall OC's on top of that! Then they'll spread the word that these are reliably the best OC'ers overall, and then these cards will fly off the shelves!!!"
And then the people who're actually in charge of these things will look at the increased costs/lowered profits involved and go "Yeah ... no. Our marketing people have told us we'll sell out all 2000 anyways ... sorry, not worth it"
Edited by brettjv - 5/5/11 at 1:45pm