Well, if that's a genuine statement - performance gains are absolutely still possible on 28nm, and Maxwell does have some interesting architectural enhancements that make it quite different than anything that has proceeded it. Case in point, Apple's A7 mobile SOC has basically blown away everything else in terms of the ARM SOC race despite being on the same lithography process, so there's definitely room for improvement. The only thing that cannot improve is density, but that may not be a big deal when considering the trade offs. 28nm's trade off is that it is mature and reasonable in price - 20nm will be extremely expensive, I know I saw this figure somewhere before but the "break even" point for any company using 20nm from the get-go will be something like 500-750k units. That is quite a doozy and very much more expensive than 28nm.
As well, you must keep in mind that TSMC's marketing always makes everything sound rosy when they describe their process enhancements. The facts are, TSMC has never been truthful in their transitions, their 20nm will be considerably worse than intel's 22nm with FinFETs, TSMC is still in 20nm "risk production" ie it is NOT READY, and TSMC's risk production for 28nm took nearly 1.5 years while they worked out the kinks. Besides which, even their board of directors basically stated that May 2014 is when volume production starts (if you believe them, they had VERY early esitmates for 28nm as well) so that means if volume production starts then, products ship in 2H 2014. And those products will be incredibly expensive.
It is what it is. This will only get worse with time. Videocardz.com had an article regarding Maxwell sometime a couple of weeks ago and the word was, Maxwell is being introduced at 28nm and will be refreshed on 20nm when the process is mature. I think that personally is fine and wouldn't read into it too much - gains are still *very much* possible on 28nm.
Personally, I think Apple and Qualcomm will be fighting over 20nm wafers from the outset just because they have oodles of money, way more than nvidia or AMD combined. From there, it will take some time for the other players to catch up. That's just the way I see it. Certainly nvidia is in a better position than AMD to pay for wafer costs, but it won't be easy for NV by any means IMHO. We'll see what happens in any case.
If i'm wrong, and I hope I am, I would welcome it. But I can't see 20nm being a welcome sight for anyone if they do somehow show up in NV GPUs, because the wafer costs are pretty ridiculous. Then again, NV does have the ability to charge more for newer technology? Again, we'll see what happens.
Facebook
Twitter