Intel's 10nm 'Cannonlake' delayed, replaced by 14nm 'Kaby Lake'

JoshuaHem

Posts: 9   +0

Intel has led the desktop processor market for the last decade, dominating in price and performance nearly without challenge since the introduction of the Core 2 chips in 2006. However, as of late the challenges of improving silicon chips has slowed their progress.

Combined with a lack of true competition from their rival AMD, PC enthusiasts have been left somewhat wanting in terms of faster and more powerful processors. Today, enthusiasts have more bad news: Intel's Cannonlake, originally slated for 2016 and later 2017, is now on hold indefinitely. Instead, Intel's current processors--of the Skylake platform--will be succeeded by the less drastic Kaby Lake platform.

Intel's efforts follow a plan known as the 'Tick-Tock' model. Every 'Tick' represents a shrinking of the previous processor design to a smaller manufacturing process. This results in greater power savings, slightly improved performance, and overall a refining of the previous design.

Every 'Tock' represents a leap forward in design, introducing a completely new processor architecture. This usually results in much greater processing power improvements and new features. Essentially, things get smaller every Tick, and a new design is created every Tock. Skylake, the current new release from Intel, is the Tock, a brand new design on the already-established 14nm manufacturing process. Cannonlake was to be a Tick, shrinking the design down to 10nm.

Little is yet known about Kaby Lake. It will be a 14nm process, but whether it will take the place of a Tick on Intel's roadmap or a Tock remains to be seen. Kaby Lake processors are expected to contain the standard dual and quad core offerings. There will be five lines of processors, Kaby Lake Y, U, H, and two S lines.

The desktop offerings, the S line, will utilize the same socket as the current Skylake processors, the LGA1151, with simple drop-in replacement support. Design power will be between 4.5 watts and 91 watts.

Kaby Lake represents an uncomfortable truth in the silicon processor world as of late: manufacturing is becoming more and more difficult. Cannonlake and the 10nm process has been delayed several times. Enthusiasts may note that progress is so slow, even venerable Sandybridge processors such as the 2500k and 2600k, launched in 2011, hold their own against and occasionally surpass modern chips in terms of raw performance.

Though the dimensions of the manufacturing node now bear little relation to the actual physical size of features on the chip, designers are now running up against atomic limitations in size and performance.

Moving beyond 14nm has been thought to require serious rethinking in design, and if Kaby Lake is any indication, that may prove to be true. The age of rapid miniaturization may soon come to an end. It may take new, radical innovation to continue moving forward. Only time will tell if designers and manufacturers are up to the task.

Permalink to story.

 
As bad as it is for the advancement of technology, at least these delays gives everyone else time to catch up with Intel on manufacturing processes and increase competition, which is good for the consumers. And when the next tech (post silicon) comes, hopefully everyone starts in equal footing again.
Any word of how the silicon replacements and memristors development is going?
 
14nm is no cake-walk in a mass-production FAB... even with 3D transistors where you have less leakage, its still not that easy with that extra process margin to reliable hit yields that make it a home run. So basically, <10nm with even smaller gates is not as close as a lot of people think it is. Sure, maybe a few wafers here and there, but not anywhere near ready for volume production I would say.
 
As bad as it is for the advancement of technology, at least these delays gives everyone else time to catch up with Intel on manufacturing processes and increase competition, which is good for the consumers. And when the next tech (post silicon) comes, hopefully everyone starts in equal footing again.
Any word of how the silicon replacements and memristors development is going?

Any word of how the silicon replacements...

not even close, 2022+ I would say and that's being aggressive. Reason is a modern semicon fab would have to be purpose built for it and that's in the 4B+ range for an LSI fab...
 
My 2500K keeps up? Cool, but too bad none of the newer features I've been looking at on motherboards (particularly M.2) are on 1155 boards (which they do not make new models anymore).

I do intend on a completely new system by the end of the year, I wonder what platform I'll use...
 
"dominating in price and performance"

I call BS on that. Yes, Intel has the more powerful CPUs. If you want/need the fastest CPU out there get an Intel - but you'll pay for it. Looking at price/performance at lower performance levels (which are more than enough for almost all users), and looking at the entire system, AMD makes sense and is a good buy.
 
"Enthusiasts may note that progress is so slow"
Enthusiasts do not dictate the pace at which technology develops. Multi billion-dollar budgets do. Unfortunately, you can only "throw" so much money at a problem. As manufacturing size shrinks ever smaller, development costs and failure rates skyrocket. Would Enthusiasts be willing to pay $1000 for a 7nm quad core? How about $4000 for a 5nm quad core? They probably can. But product sales from few million enthusiasts cannot pay for the multi billion-dollar budget spent to develop those chips. Which is why the world needs to move on from the concept of "performance gains" from die shrink. Alternative materials and new architectures are the way to (profitably) manufacture the next generation of processors.

"I call BS on that. Yes, Intel has the more powerful CPUs ... AMD makes sense and is a good buy."
What a good buy "is" is relative. The bottom line is profitability. The company which has better sales, cash flow and stock value will produce the best processors. A more profitable company has more R&D budget. To the mind of an average consumer a few dollars is a deal breaker. If such a thing is such a good way to evaluate electronics, why isn't AMD's sales, cash flow, stock value and R&D budget getting larger at the same rate Intel's is?
 
What's with Intel and all their "Lake" stuff? Are they struggling for better sounding nomenclatures?
They always use their suffixes for more than one generation to point out the updates. Like their "dale" stuff (Allendale > Wolfdale > Clarkdale), their "field" stuff (Kentsfield > Yorkfield > Bloomfield/Lynnfield), their "bridge" stuff (Sandy Bridge > Ivy Bridge) and their "well" stuff (Haswell > Broadwell). The only chip name that wasn't part of this was "Conroe", but it was just a different version of Allendale.
 
Umm no. This is a new chip not a frequency bump and TIM change, get it right.
Really? I doubt it. How will that chip for this "refresh" platform be any different than skylake? It's just like a haswell refresh (with h/z97 series chipset) (so yeah, "devil's canyon" might not be the right association)
 
Profitability is dead. the time to serve our higher instincts for the next big thing has come. profitability's long pole has been to continue on the same path (reduce costs). now that biotech and algorithms are out-competing moore's law how long will it be before we eschew profitability for innovation and bring transformative insight to the quantum computing field instead of trying to fit the square quantum peg into the electromagnetic round hole?

it's been an amazingly peaceful ride these past few decades. it's time to start getting dirty again and provide support to some crazy people to try out their ideas. even entrepreneurship has barriers to entry that blocks innovation. at least that was how we used to do things, before we forgot that there are no rewards without risk.
 
Really? I doubt it. How will that chip for this "refresh" platform be any different than skylake? It's just like a haswell refresh (with h/z97 series chipset) (so yeah, "devil's canyon" might not be the right association)

They're planning far enough in advance go for another 14nm design, instead of a 10nm, that they can still add new instructions, and maybe introduce more efficient implementations of other recently-introduced instruction sets. (Like transactional memory, if they have any good ideas that didn't make it into Skylake.)

I'm hoping for AVX-512 in desktop/laptop Kaby Lake. (The current roadmap is for AVX-512 support to be released in Skylake Xeon, but not other Skylake versions.)

-- Peter Cordes
 
Profitability is dead. the time to serve our higher instincts for the next big thing has come. profitability's long pole has been to continue on the same path (reduce costs). now that biotech and algorithms are out-competing moore's law how long will it be before we eschew profitability for innovation and bring transformative insight to the quantum computing field instead of trying to fit the square quantum peg into the electromagnetic round hole?

it's been an amazingly peaceful ride these past few decades. it's time to start getting dirty again and provide support to some crazy people to try out their ideas. even entrepreneurship has barriers to entry that blocks innovation. at least that was how we used to do things, before we forgot that there are no rewards without risk.

But the profit motif and free enterprise competition are the driving forces behind innovation.

"Even entrepreneurship has barriers of entry that blocks innovation" Yes! The government plays favorites with big corporations over smaller business, the result being lots of red tape that acts as a barrier to entry thus stiffing free competition.
 
Back