Lat­est tech­nol­ogy news

It’s time to add ‘Can­non Lake’ and ‘Cof­fee Lake’ to your vo­cab­u­lary, writes Mark Hachman

Tech Advisor - - Contents -

Nor­mally, In­tel launches one new mi­cro­pro­ces­sor every year. But the firm is qui­etly sig­nalling that PCs shipped dur­ing the hol­i­day 2017 sea­son could fea­ture one of two new de­signs, us­ing ei­ther its ag­ing 14nm process or the up­com­ing 10nm tech­nol­ogy.

The two new names to add to your In­tel vo­cab­u­lary include Can­non Lake, a 10nm chip that In­tel showed off at CES, and an un­named fourth-gen­er­a­tion 14nm chip that some are re­fer­ring to as Cof­fee Lake. The lat­ter chip is per­haps the most in­ter­est­ing, as it would ex­tend In­tel’s 14nm chips to a record four gen­er­a­tions, fol­low­ing Broad­well, Sky­lake, and the cur­rent Kaby Lake chip.

When In­tel added a third 14nm chip 2015, sig­nalling the end of the com­pany’s vaunted ‘tick-tock’ man­u­fac­tur­ing process, the in­dus­try freaked out. A fourth, then, would be ab­so­lutely un­prece­dented. But In­tel al­ready seems to be set­ting the stage to con­vince in­vestors and cus­tomers that it can man­u­fac­ture four straight gen­er­a­tions of chips on the same 14nm process with­out giv­ing up con­sis­tent per­for­mance im­prove­ments, which is what cus­tomers care about, af­ter all.

At its re­cent in­vestor day, In­tel said that its eighth-gen­er­a­tion Core chips, which should include Can­non Lake and Cof­fee Lake, will de­liver a 15 per­cent per­for­mance im­prove­ment over Kaby Lake. That’s good enough for most con­sumers. But en­thu­si­asts have to won­der where the ad­di­tional per­for­mance will come from, and whether it will open the door to AMD’s own ri­val, Ryzen. There’s also a broader question: if In­tel’s be­ing forced into ship­ping a fourth 14nm chip, will the same sce­nario play out within the 10nm gen­er­a­tion?

Here’s how 2017 shapes up

In­tel hasn’t made a big deal of the fact that it’s pur­su­ing par­al­lel 10- and 14nm

strate­gies, but both chips have been fi­nally re­ferred to pub­licly. At CES, In­tel showed off a two-in-one run­ning Can­non Lake, later claim­ing that it would ship Can­non Lake chips in 2017. Its plan for a sep­a­rate 14nm chip ap­peared in a roadmap slide dur­ing its re­cent in­vestor’s day, ti­tled ‘Ad­vanc­ing Moore’s Law on 14nm’. That roadmap in­cluded an il­lus­tra­tion show­ing that the com­pany ten­ta­tively plans to ship those 14nm, eighth-gen­er­a­tion Core chips dur­ing the sec­ond half of 2017, too.

Per­haps the most in­ter­est­ing as­pect of what some are call­ing a ‘split’ approach is how Cof­fee Lake and Can­non Lake are po­si­tioned. In­tel is be­lieved to be tar­get­ing Can­non Lake at more pre­mium niche mar­kets such as low-power ul­tra­books, leav­ing the well-es­tab­lished 14nm process pow­er­ing Cof­fee Lake for main­stream note­books and desk­tops un­til the 10nm Can­non Lake comes fully up to speed. That strat­egy avoids forc­ing In­tel to bet ev­ery­thing on the un­proven 10nm process, and al­lows them to fo­cus on pre­mium, high­mar­gin prod­ucts, an­a­lyst Dean McCar­ron said—the strat­egy that helped In­tel’s client group grow op­er­at­ing prof­its by 30 per­cent even as PC ship­ments de­clined in 2016.

Gen­er­ally speak­ing, here’s how 2017 could shape up in terms of CPU re­leases: In­tel’s Kaby Lake holds sway un­til March, when AMD has said that the Ryzen chip will launch. Some­time near the sum­mer – late May, if his­tory holds – In­tel should re­lease its high­end Sky­lake-E chips for gam­ing PCs. AMD will ship its Raven Ridge chip for lap­tops and some desk­tops in the au­tumn, which is also roughly the time frame to ex­pect In­tel’s Can­non Lake and Cof­fee Lake.

How Kaby Lake com­pares to Ryzen will play an enor­mous role in AMD’s fu­ture, but ques­tions will quickly be raised about how the 14nm chip, Cof­fee Lake, will shape up, too. You may al­ready be ask­ing this question: if Cof­fee Lake is In­tel’s fourth chip on the same 14nm process, how can it pos­si­bly be any faster than its pre­de­ces­sors?

New tech­nol­ogy

His­tor­i­cally, In­tel in­creases its chip per­for­mance from prod­uct to prod­uct in two ways: de­sign and man­u­fac­tur­ing. This is the afore­men­tioned tick-tock process; a new change to a chip’s de­sign (tick) is fol­lowed by a ‘process shrink’ to an im­proved man­u­fac­tur­ing process (tock). Both meth­ods gen­er­ate per­for­mance im­prove­ments. The ‘tock’ is where Moore’s Law en­ters: the num­ber of tran­sis­tors in a given area dou­bles every 12- to 18 months as new man­u­fac­tur­ing tech­nolo­gies are de­vel­oped, giv­ing chips a ‘free’ per­for­mance boost. But given that In­tel is leav­ing its man­u­fac­tur­ing process un­changed, doesn’t that mean that its ‘tock’ im­prove­ments will dis­ap­pear?

Not ex­actly. Over time, both the tick­tock model as well as what it de­scribes has bro­ken down. Dur­ing the Sky­lake and Kaby Lake gen­er­a­tions, In­tel squashed the ‘tick’ and ‘tock’ to­gether some­what, tweak­ing both the de­sign and the process tech­nol­ogy si­mul­ta­ne­ously.

Even a de­scrip­tive term such as ‘14nm’ means has also be­come more vague. Sources con­firmed a tweet by The Mot­ley Fool’s Ashraf Eassa, that In­tel is us­ing a man­u­fac­tur­ing short­hand to dis­tin­guish

be­tween Broad­well/Sky­lake (14nm), Kaby Lake (14nm+), and Cof­fee Lake (14nm++). That, plus In­tel’s shift to a more three­d­i­men­sional FinFET ar­chi­tec­ture, has made identifying each gen­er­a­tion by a gate length even more out­dated, added Pa­trick Moor­head, a for­mer AMD fel­low and now an in­de­pen­dent an­a­lyst with Moor In­sights and Strat­egy.

“I ac­tu­ally be­lieve that In­tel’s 14nm+ is ac­tu­ally closer to other peo­ple’s 10nm than they get credit for,” Moor­head ar­gued.

In his in­vestor day pre­sen­ta­tion, Murthy Ren­duch­in­tala, the head of In­tel’s client chip and IoT busi­ness, used a dif­fer­ent met­ric in a bid to clear up the con­fu­sion once and for all: logic cell area. Un­der that met­ric, he said, In­tel has steadily shrunk the logic cell by 50 per­cent over every two years, and has main­tained a steady three-year lead over its ri­vals.

All of this may sound hor­ri­bly ab­stract, but the bot­tom line is this: Moore’s Law has driven the en­tire com­put­ing in­dus­try for lit­er­ally decades. It seems to be slow­ing down, and in some ways, it is. But Moore’s Law is no longer as ac­cu­rate as it once was, and that in­ac­cu­racy masks some of the progress In­tel’s mak­ing.

No easy an­swer to im­proved per­for­mance

Nev­er­the­less, an­a­lysts ex­pect that In­tel will turn other knobs when im­prov­ing the per­for­mance of a fourth-gen­er­a­tion 14nm chip. Since In­tel’s roadmap slide con­firms that its eighth-gen­er­a­tion Core chip runs 15 per­cent faster than the cur­rent Kaby Lake, that im­plies that Cof­fee Lake will fea­ture a re­vised de­sign, an­a­lysts say, sim­i­lar to how the 14nm Kaby Lake chip im­proves over its 14nm pre­de­ces­sor, Sky­lake.

The “cheap, dirty way” to im­prove per­for­mance in­volves adding cache mem­ory, McCar­ron ex­plained. “Putting more cache in al­ways solves the prob­lem.”

What’s more likely, though, is that In­tel will sim­ply com­bine a num­ber of small im­prove­ments, McCar­ron said, such as run­ning core clocks slightly higher and ad­just­ing power. For his part, Moor­head said that he be­lieved In­tel would bet more heav­ily on im­prov­ing the in­te­grated graph­ics en­gine. “My gut in­stinct says that it’s a lot of GPU, a lit­tle bit of clock [speed], and po­ten­tially not a whole lot more in­struc­tions per clock” to achieve that 15 per­cent im­prove­ment, he added. Long gone, then, are the days when In­tel di­alled up the clock speed to dra­mat­i­cally in­crease per­for­mance. As Moore’s Law slows, the firm will likely take a lit­tle bit here, and a lit­tle bit there, to grad­u­ally push per­for­mance higher.

So which chip should you buy, Cof­fee Lake or Can­non Lake? Be­cause of the way In­tel plans to re­port­edly split up the two ar­chi­tec­tures, nei­ther should com­pete with each di­rectly – at least at first. Re­mem­ber, In­tel’s de­lib­er­ately de­lay­ing its 10nm process as part of its ‘data cen­tre first’ strat­egy, giv­ing the 10nm tech a chance to ripen on the vine. Even­tu­ally, it seems like Can­non Lake and Cof­fee Lake col­lide, and then what? At this point, we sim­ply don’t know.

It cer­tainly looks like this could be the sum­mer of Ryzen. How quickly In­tel can steal its thun­der with Can­non Lake or Cof­fee Lake re­mains to be seen

In­tel re­vealed its plans to ex­tend 14nm tech­nol­ogy through the eighth-gen­er­a­tion Core prod­ucts at its in­vestor con­fer­ence

In­tel chief ex­ec­u­tive Brian Krzanich shows off Can­non Lake hard­ware at its CES 2017 pre­sen­ta­tion

In­tel has been qui­etly work­ing to im­prove its Iris Pro in­te­grated graph­ics, and we’d ex­pect that trend to con­tinue

This is the met­ric In­tel is now push­ing to show how its man­u­fac­tur­ing is con­sis­tently im­prov­ing: logic cell scal­ing

Newspapers in English

Newspapers from UK

© PressReader. All rights reserved.