By using this site, you agree to our Privacy Policy and our Terms of Use. Close
Soleron on 16 September 2011
Edit Thread

haxxiy said:
Unfortunately, I knew it was fake the instant I read about XDR2. AMD would shot themselves in the feet by making such a deal with Rambus. Oh, and there will be no 28nm high-end parts this year,

AMD promised them this year and demoed a working GPU. It depends on whether TSMC can improve the process in time. I'm betting no but it is purely a 28nm issue and not an AMD one.

As a sidenote, it would be not that great of an improvement even if it were real TBH, which ironically made them a bit more reliable even dreaming about XDR2. Some hardware areas - if not most of them - are crawling compared to how fast they advanced in the late 90's to mid 00's. There is simply not much demand for more, outside of portables, and some obvious physical restrictions.

Yes. Memory bandwidth is not something consumer applications of any kind are short of; look how little DDR3, tri-channel memory or PCIe-2.0 did. The two key metrics consumers benefit from are battery life and disk performance (i.e. SSDs).

Moore's law is going to shit and most manufactures, mainly Intel, do not admit so, in order to please shareholders and blind them to the obvious issues they are going to face very soon. Do not be surprised to see hardware manufactures defending some kind of killswitch the way developers are moaning about used games, a few years from now.

Funny you should say that; Intel announced two days ago that scaling is dead. For real.

http://download.intel.com/newsroom/kits/idf/2011_fall/pdfs/2011_IDF_Otellini_Opening_Keynote.pdf

Page 11.

But before it is dead it will become very expensive. Many companies already stopped in-house process nodes (notably TI and AMD). Several foundries stopped bothering to develop new nodes (Chartered and UMC).

Currently we only have Intel, Globalfoundries/IBM/Samsung and TSMC doing development. I think IBM and Samsung won't bother to go to 16nm on actual production and TSMC will give up on 11nm due to the exponentially rising costs.

Intel and Globalfoundries will be the last companies standing when scaling stops.