As far as the Samsung/GloFo licensing goes, it would make sense that GloFo had customer(s) lined up before they licensed Samsung's tech. That could be AMD, but there's been nothing but pure speculation on that. Not even rumors.
GloFo have not been doing very well up until so far. By joining up with Samsung they are doing themselves a big favor, while Samsung can invest less on R&D than they otherwise would have and gains capacity by sharing technology.
And Samsung's already doing pretty pioneering work:
http://www.extremetech.com/extreme/...t-the-wonder-material-into-real-world-devices
As for AMD, it's well-known that they have very good deals with GloFo, in large part because the two companies have a history together. I'm expecting them to stay with TSMC for some time to come. Switching fabs is a cumbersome and timeconsuming process. But if it is done, it would probably be for the 2016 GPUs on 16/14 nm.
Just because it isn't coming on 28nm (assuming SweClockers is right) doesn't mean it's coming on 20nm. Don't get me wrong, I'm not saying this. It's just other rumors that are floating around.
I mean, everything is just rumors at this point. But it is also important not to sink into a kind of relativism where we judge everything by the same standards. We need to differentiate between sites that jump at everything, á la Wccftech or Videocardz, and sites that are much more selective and have a good track record, like Sweclockers, the german Computerbase and other websites. We should still keep in mind that rumors are, by the end of the day, rumors, but not all sources are equal, which was the point I was making. If someone wants to make a contrary point, feel free, but have a credible source with a good track record to do so.
If I was going to go out on a limb to the pro-28 nm Big Maxwell-crowd, then I guess we could see a gimped Big Maxwell on 28 nm late this year/early next year that gets a significant boost both from more SMM's activated and a more power efficient node in late 2015. So in other words, Big Maxwell is coming on 20 nm regardless. But in this alterate scenario it could come in two waves. I'm not saying this is what I believe, I personally believe we will see Big Maxwell on 20 nm only, but I'm saying that could be a model.
Where the theory of no big Maxwell on 20 nm goes out the window is that these people have to tell a compelling story of how Nvidia is going to play its card next.
Pascal is
at least 2 years away, could be two and a half years away if it is launching in H2 2016. So say you launch big Maxwell on 28 nm this year or early next year. What are you going to do during most if not all of 2015 and 2016? No new architecture is coming up. Are these people saying we'll be seeing 16 finfet(
true 16 nm finfet) in late 2015? Considering that 20 nm has seen endless delays, those odds crawl close to zero.
Neither the architecture nor the node card would be at play if we don't see big Maxwell on 20 nm until 2-2.5 years.
Of course Nvidia is saving 2015 for 20 nm for big Maxwell, either as a fresh offering or as a "full release" where all SMMs are activated from a previously gimped 28 nm version. And the reason is that that is the only card they can play since Pascal's out in 2016 and if 20 nm isn't available, according to the logic of these people, the odds that Nvidia is going for true 16 nm finfet is basically nil unless they switch fabs to 14 nm, but as I said before, changing fabs takes plenty of time and not something you do in a few months. That'd be for 2016+ (if at all).
Facebook
Twitter