Xbit: Next-Generation DDR4 Memory to Reach 4.266GHz - Report.

[richardginn]

http://www.xbitlabs.com/news/memory/...Hz_Report.html

Faster is better

[kpxgq]

isnt this backwards since we have GDDR5?

[James]

isnt this backwards since we have GDDR5?

Nope. G-DDR5 and DDR4 are different.

The Graphics variants of DDR tend not to lend themselves well to high capacity chips.

[drdoom]

Doesn't look like it will be anytime soon. I'm still on DDR2.

[Nater]

Yeah, I don't think this will be anytime soon. If I had to hazard a guess, maybe intel will be using it by the 22nm node, but 16nm seems more likely to me. I think the first you'll possibly see it is Haswell.

[drdoom]

Sorry off the subject but what is the theoretic limit and practical limit to die size?

[Nater]

To die size? Yes. Technically, I don't know what the theoretical limits are, but the economics of the thing limit them to 700mm2. That is the size of the current Itanium 2s, the POWER 7 would be around the same size except they went to using eDRAM for L3 cache instead of SRAM. nVidia's GF100 chip is about 522mm2.

As far as process size goes, they really haven't run into a hard limit yet for processors, yet. NAND flash is supposed to run into serious quantum tunneling problems under 22nm. I wouldn't be surprised if they're able to push it below that, but there is a reason than intel is working very hard on phase-change memory. It can scale much more easily down into the realm of nanoelectronics, >16nm.

45nm bulk CMOS doesn't have much advantage over 65nm bulk CMOS unless you use hi-k metal gates and other assorted technologies. It might be different for the silicon on insulator process that AMD uses, but intel's 45nm processors were generally more power efficient than AMD's. AMD is going to have to go to hi-k for their 32nm node, so the ground they gained on intel at 45nm by avoiding hi-k is already lost.

Current deep ultraviolet lithography is going to run out of steam much below 13.5nm, fabs will then have to switch to extreme ultraviolet lithography. Intel has been developing it for more than a decade with little success. Below 16nm is probably where and when microprocessors especially will move away from silicon and towards carbon (graphene). It's unlikely to happen all at once, but it will start to happen.