Intel breaks ticktock

The magic smoke.

Moderators: phlip, Moderators General, Prelates

User avatar
Zamfir
I built a novelty castle, the irony was lost on some.
Posts: 7303
Joined: Wed Aug 27, 2008 2:43 pm UTC
Location: Nederland

Intel breaks ticktock

Postby Zamfir » Sat Aug 08, 2015 7:57 pm UTC

http://www.anandtech.com/show/9447/intel-10nm-and-kaby-lake

A few weeks ago already, but I missed it. Intel announced a change in plans, adding a third processor design on the 14nm node. And intel takes its 2-year ticktock cycle religiously serious. It's not an enormous surprise, their schedule has been slipping for a while. But now they made it formal: for the next years, intel won't stick to Moore's law.

EUV seems to be getting somewhere, but it's still not production ready and the date of readiness keeps slipping into the future. It's now planned for 2019 and 7nm. At that point, intel hopes to return to ticktocks.

wumpus
Posts: 500
Joined: Thu Feb 21, 2008 12:16 am UTC

Re: Intel breaks ticktock

Postby wumpus » Sun Aug 09, 2015 4:24 pm UTC

Are all the benefits from 14nm (really anything below 28nm) just power efficiency?

It seems that since about 32nm (Sandy Bridge), the only improvements to desktop power has been in AVX[2] and specialty instructions, possibly with a little bump for DDR4. GPUs have stuck with 28nm (AMD released 28nm 7970 in Jan, 2012. Nvidia unleashed the 28nm 680 in March, 2012). 3D flash lets flash "regress" to 32nm (although it obeys Moore's law since it packs more transistors cheaper than before), although the new Intel/Micron memory is said to handle more advanced processes better than flash.

I assume that memory is still moving to smaller pitch fabs (google hits list 20nm). Of course smartphones are a huge gaping market for RAM (a gig or two per phone, and low power memory at that). Also I assume that servers haven't lost their endless lust for RAM, and it would likely make sense to use low power RAM at that.

User avatar
Zamfir
I built a novelty castle, the irony was lost on some.
Posts: 7303
Joined: Wed Aug 27, 2008 2:43 pm UTC
Location: Nederland

Re: Intel breaks ticktock

Postby Zamfir » Sun Aug 09, 2015 7:44 pm UTC

Yeah, the smaller transistors don't seem to get much faster, and rumours say the last shrinks have not been cheaper (per transistor) either. They might be getting more expensive per transistor.

So the remaining gains are lower power, and lower integration costs. Even if one die at the smaller node costs the same as two dies at the larger node, device makers are still happier to have more functionality in one place. Like Qualcomm's modems.

Apparently, Intel is expecting problems to get to 10nm before they fully tried, just extrapolating from the problems at 22 and 14. Given the amount of money and expertise Intel can throw at it, and the prestige they have tied up in shrinks, the rest of the world will presumably slow down their shrinks as well.

And there will be knock-on effects. If the processors don't move to smaller sizes, then their factories won't become available for other products.

User avatar
Link
Posts: 1327
Joined: Sat Mar 07, 2009 11:33 am UTC
Location: ᘝᓄᘈᖉᐣ
Contact:

Re: Intel breaks ticktock

Postby Link » Mon Aug 10, 2015 10:52 am UTC

Well, IBM is working on 7 nm technology now, so there's some room for improvement. That said, I really think we're approaching the limit of what can be done with conventional CPUs. There are plenty of possible ways around that (3D chips, memristors, spintronics, ...) but there doesn't appear to be a whole lot going on in those fields yet.

In the meantime, Intel is also working on new nonvolatile memory, so they're likely to push some boosts to computing in general even if CPUs are starting to stagnate.

User avatar
Zamfir
I built a novelty castle, the irony was lost on some.
Posts: 7303
Joined: Wed Aug 27, 2008 2:43 pm UTC
Location: Nederland

Re: Intel breaks ticktock

Postby Zamfir » Wed Aug 12, 2015 8:36 pm UTC

Yeah, the technical limits aren't reached yet. But in the end, Moore's law is based as much on economics as on technical grounds. The advantages of shrinks are so great that people poured in almost any amount of money necessary. That might end at some point. So people might say, we can do another shirnk but investments will be high, production costs will be high so not every market will switch, we'll need 5 years or 10 years to recoup the investments.


Return to “Hardware”

Who is online

Users browsing this forum: No registered users and 4 guests