Details Leak on Intel’s Upcoming Ice Lake CPU, 10nm Schedule

This site may earn affiliate commissions from the links on this page. Terms of use.

Intel has taken the unusual step of releasing (or leaking) information on the company’s next-next-generation processor. The new, 10nm chip will be codenamed Ice Lake and isn’t expected to debut until after Coffee Lake, which is expected to launch later this year on the 14nm++ process. We already knew Intel was going to reveal new information on its upcoming CPU family on the 21st (the same day as the eclipse), but it’s rare for the company to let much slip about its hardware plans more than one generation at a time.

10nm-1

Click to enlarge.

Right now, it looks as if Intel will introduce 10nm parts (codenamed Cannon Lake) only in 2018. Just as Intel brought 14nm up with low-power Broadwell chips before deploying it for other cores, Cannon Lake is likely to focus on lower-power chips first. This was actually implied by Intel’s 10nm documents, released earlier this year. We’ve put the relevant slideshow below; all the slides are relevant to Intel’s 10nm plans, but the last slide is particularly useful.

Intel told us back on March 31 that it would introduce a new set of products built on 14nm++, and that 14nm++ would actually offer higher performance than its first-generation 10nm. 14nm++ isn’t expected to be as strong a leap over 14nm+ as 14nm+ was over 14nm, but it should still deliver some significant improvements. Assuming Intel also plans to move its Skylake-X / Skylake-SP cache configurations to its Coffee Lake CPUs, we should see some intrinsic uplift there as well. Remember, the gains shown here only reflect transistor improvements, not architectural changes.

Anandtech thinks we’ll see Coffee Lake and 14nm++ as the only option on desktops, laptop parts are likely to be a mix of 14nm++ (at or above 35W), and 10nm will be reserved for 15W and lower-TDP parts. This would mimic how Intel debuted its 14nm technology and it makes sense — early in a process cycle, when yields are the worst, the best way to make certain you can minimize wafer cost is to build as many CPUs per wafer as possible. The more good chips you get per wafer, the more CPUs you have to absorb the cost of low-yield production.

As for why Intel’s 10nm timeline has slipped so much as to necessitate pushing 10nm back, Ars Technica reports hearing that the company had to reassign engineers from its 10nm process to fix issues it was having with 14nm. This took long enough that the company couldn’t make up the difference and hit its original launch targets with 10nm. This isn’t particularly surprising, either, though it wouldn’t surprise us if Intel adopted a different tactic for its own process nodes going forward.

Right now, the foundries and Intel have pursued very different strategies for targeting new nodes. Samsung and GlobalFoundries are both taking an incrementalist approach. Their 10nm technology is similar in key metrics to Intel’s 14nm tech and their 7nm technology, when it arrives, will be similar to Intel’s 10nm tech. GlobalFoundries hopes to have its first-generation 10nm technology in high volume manufacturing by the second half of 2018, though no firm introduction date has been given. It’s not unusual for customer shipments to lag four to six months behind the introduction of high volume manufacturing, depending on which chips are being built and how the process node launch aligns with the customer’s product launch schedule. TSMC, for example, started its high volume 10nm manufacturing at the beginning of the year but those chips didn’t hit the market until June.

It looks right now as if the foundries may have made up some ground on Intel, but we’ll wait and see how 7nm shapes up before drawing any firm conclusions there. If they have, Intel may well move to a new model for its own parts, one that simplifies the update process by taking smaller steps at once.

It’s also all too likely that delayed EUV rollouts or trouble with the 7nm node could delay Samsung, TSMC, or GlobalFoundries. It’s getting harder and harder to hit new nodes, and every foundry has problems from time to time. TSMC had a terrible 40nm deployment, GlobalFoundries had to cancel its own 14nm technology and adopt Samsung’s, and Samsung…well, Samsung hasn’t really had any cataclysms, but its first implementation of big.Little was a hot mess and it’s only been fairly recently that the company started pushing to expand its foundry business. Even when it built chips for Apple, Samsung’s foundry business mostly served its own needs first. Give it enough time, and Samsung will trip as well.

Let’s block ads! (Why?)

ExtremeTechExtremeTech