AMD and Globalfoundries to Make 28nm Parts
In a recent interview, Globalfoundries vice president of manufacturing systems and technology Tom Sonderman said that the company plans to make GPUs using 28nm processes in the near future.
In a recent interview, Globalfoundries vice president of manufacturing systems and technology Tom Sonderman said that the company plans to make GPUs using 28nm processes in the near future.
Recently X-bit Labs published an interview with AMD's Tom Sonderman, discussing AMD's spin-off company (or rather the former manufacturing division), Globalfoundries, and its future corporate plans. The four-page interview offers a wealth of knowledge, discussing topics such as the Globalfoundries name, the origins of the company, and the advantages the company will offer in comparison to other existing foundries. However, what stood out in the interview was Sonderman's comment towards the end, when X-Bit Labs asked about AMD outsourcing its chipset and graphics chip production to Globalfoundries.
"We intend on competing for AMD’s graphics business in the 32nm/28nm technology node," he said. However, as X-Bit pointed out, AMD is obliged to require both microprocessors and GPUs from Globalfoundries, as AMD is one of the owners; using the term "competing" may have been inaccurate. Still, Sonderman's answer hints to the possibility that the company may be ready for 28nm processing soon.
Additionally, Sonderman said that the company is already sniffing around for additional clients outside AMD. "We are actively engaging with multiple tier 1 fabless and fablite semiconductor companies who require leading-edge capacity," he added. "We’re not ready to announce any new relationships yet."
Could his comment mean that companies such as Nvidia, Nintendo, and Microsoft are potential clients? "PC platform technologies (CPUs, GPUs), wireless, game consoles, and telecom are a few examples of markets we’re targeting," he confirmed.
Sonderman also indicated that 32nm bulk silicon is already running in Dresden, and will be ready to accept customer designs later this year; it will even offer an aggressive production ramp in 2010. However, he did not offer additional details regarding to 28nm processing, and when the company plans to implement the technology. Stay tuned as Globalfoundries reveals more in the near future.
Anyone knows if AMD uses plain silicon chips,or if it follows intel in using the high Metal-K process?
But the more important thing to Ati is that it has alternatives when making new GPU's. If one source is in thouble, the other may provide and ansver. I hope that this means no more problems to Nvidia. We still need at least two GPU companies in the market.
It would be nice to see both ATI and Nvidia to buy their production from Globalfoundries... Not likely, but interestin thing to see ;-)
Moore hasn't let us down yet but the physical material limitations are tricky...tricky along the lines of trying to make a room temperature super conductor. Not necessarily impossible but not really doable either with current knowledge.
Of course...process shrinking isn't the only way to improve performance either.
I wonder what the odds are of nVidia chips getting produced by Global. It makes business sense but it would be darn ironic.
It's more like a sudden fried CPU than a slowly hotter getting cpu.
Perhaps even a more instable processor, one that at times can perform faster than at other times,and depending from processor to processor.
Up 'till now processors where more consistent.
Overclocking on a certain model could be reproduced at home with a similar cpu upto a few percent of the end results. But the smaller we get, the more of that random factor we'll see.
20nm is excellent for mini netbooks.
The lines are starting to blur a bit between GPU/CPU steadily as well, especially if you consider how they started vs. where they are now. With nVidia wanting GPUs to take on more processor load, Intel wanting CPUs to take on more graphics load, and ATI wanting to blend the two into a more or less seamless unit.
So it might come down to a marrying of some of the functionality of GPU/CPUs with a QPU to offload appropriate bits of work. Fun to speculate about anyway.
Nope....
Intel uses bulk silicon, the same thing they are talking about in this article. High-k just refers to the transistor gate material. They also use low-k materials. High-k isnt a type of silicon, its just talking about one small piece of the chip.
AMD uses SOI, which is different from bulk silicon. They also use both high and low k materials on their chip. They aren't using hafnium yet, which is intels current transistor gate material on the 45nm note, but they do plan to use it at the 32nm node i believe.
There is no such thing as Ge oxide for high K as it is does not a high dielectric constant ('k'), the poster is confusing the use of Ge for strain purposes (which Intel pioneered back on 90nm) and high K.
Also don't confuse 'Si running' with ready for production... to put things in perspective Intel has 22nm Si in their development fabs, but it is nowhere near ready for release.
Finally 28nm is a 'simple' optical shrink (what some refer to as a 'half-node'); this is not really a big deal from a performance perspective, but helps with cost. This typically is cranking down on litho a bit tighter while tuning some of the other process steps.
And it was not a matter of AMD 'not needing' high K for 45nm - it just wasn't ready in time, and it is not practical to implement this in the middle of a tech node due to the complexity and risk (despite AMD's claims that it 'was an option for later 45nm'). If AMD were able to implement it on 45nm they would have.
Whenever this guy posts an article, I get a sense of deja vu.
imagine having 2 of such cards in a system. play a driving game with 5 gpu's handling realistic physics and the last 3 handling realistic graphics on an ordinary 22" screen! You'd end up with a driving simulator that hamilton could use!
"NTT Japan has successfully tested a fiber optic cable...that pushes 14 trillion bits per second down a single strand of fiber." -Did you know 3.0 (youtube)
This is a type of bussiness that if you already have it, then you woudn't talk about it until release. So Intel probobly is far ahead of GF since they have about $2 Billion in RND/Year.
What's that, 2010 or 2016? If I'm a 14yo then for me Near future is about 5 minutes, and if I'm 50 then it could be 10 years. That's aclever statement when you have nothing to release.
That would give Intel a serious run for the money.