Thursday, September 20th 2018
Intel to Move Select Chipset Fabrication Back to 22nm in Wake of 14 nm Silicon Constraints
Things seem to be taking turns to the worse at Intel in accordance to Murphy's law. Not only was the company hit with a multitude of security flaws embedded in their CPUs, which puts their michroarchitecture design chops in jeopardy, but now they also have to contend with silicon fabrication snags. That Intel's 14 nm fabs are being hit with overwhelming demand for their output capacity is already a known quantity, with rising prices of Intel mainstream CPUs and reports of the company outsourcing 14 nm chip production to TSMC in a bid to increase availability - a first since the company became vertically integrated with both design and manufacturing of their own chips.
Now, reports are coming up that Intel will be moving some of its chipsets back to the 22 nm node - namely, the H310 chipset - so as to clear production capability from the 14 nm one. As you might remember, historically, Intel's chipsets have been one silicon manufacturing generation behind their CPUs. Due to the problems in Intel's 10 nm process and constrained output of their 14 nm process, this has now become a necessity again. The new H310 chipset, which had to be architecturally revised for 22 nm (which isn't as easy a thing as one might expect) will debut in the H310C or H310 R2.0 indicator. It will be physically larger (naturally) and incur in a small power efficiency loss as well, and - get this - support Windows 7, likely via asoftware/driver solution. Motherboards with the new chipset are already moving out into the supply chain.
Sources:
ETeknix, Tom's Hardware
75 Comments on Intel to Move Select Chipset Fabrication Back to 22nm in Wake of 14 nm Silicon Constraints
Hilarious
The only logical explanation is that Intel have decided to try out 14nm lithography for chipsets on the lowest-of-the-low H310, so if something goes wrong, then replacing a $50-60 motherboard will not create as much fuss or outrage as a high number of faulty/defective high-end Z370 boards, or B360-based enterprise PCs. In most cases it may even go unnoticed, cause "cheap stuff breaks".
Well... they didn't, so we end up with this strange shift in production. It doesn't look good, but it's fairly neutral.
On the other hand, the reasons why this is happening are very positive (apart from 10nm delay, obviously). Intel is selling a lot of stuff, they are getting new contracts.
Reaction on gaming sites/forums is just hilarious.It's pretty obvious by now that everyone has issues with 7/10nm. One of the most advanced semiconductor makers has just surrendered.
Everyone's working on a solution to make this idea profitable. Most likely everyone will announced readiness at more or less the same time.
Intel's delay will be the longest simply because they were the most advanced in the early development period, i.e. were first.
And I'm known to be a Intel fanboy, so I'll also say that Intel will be the first to launch an x86 CPU. Of course apart from the existing mobile i3 available since May. :p
Intel was really able to save face and increase demand by making its most significant CPU performance improvements that I can recall in a long time. i3 is now a quad core, i5 is 6 core, i7 is 6C/12T. The whole stack got 2 more cores. Before that, the refreshes were pretty anemic, making upgrades largely unnecessary.
David Kanter wrote a pretty nice article about it a few months ago, if you have to stomach for 4 pages of fairly dry, technical reading ;).
/s
The 14nm H310 was only 6w, the 22nm might be 10w. Not enough to make any real difference, and certainly not enough to throw off power regulations.
It's obvious that they planned to move CPUs to 10nm and chipsets to 14nm. But the fact that they did the latter even when the former was delayed is somehow weird.
On the other hand, it's still only H310, so not the best selling nor most important chipset. They haven't done the same with the bestselling H/Q370 yet (both desktop and mobile), so IMO the transition to 10nm is still going relatively well (compared to what we read online ;-)).
Plus, I bet they actually sell more H310 chipsets than any others. Those cheap pre-built computers that fly off the shelves, because they are cheap, usually always use the lowest end chipset.
Worth exactly no·thing if your tools are broken!
The last couple of months really feel somewhat awkward. It's like on every new head-line 'bout Intel you're thinking »Guys, stop it, he's already dead!« … Then you have to read the next, stating that „things get worse for Intel …“ and you just asking yourself »Well, how worse it can even get anymore?!«
It feels really, really strange these days …
Did karma actually bought an Itanium CPU back then? A Larrabee-card by any chance? Or some Optane SSDs? What a bitch!
So... how does one move to an older node, then? Did they just pack up the 22nm machine and put it away? I would imagine 14nm, 22nm etc "tools" are merely part of a larger machine/process that produces these things, so it wouldn't make much sense to swap out 14nm for 22nm, unless I'm grossly misunderstanding something, which is very well possible... I'm no fab expert, nor have I ever even been near one.
ark.intel.com/products/series/126380/Intel-300-Series-Chipsets
It's a bit weird that some of the chipsets are listed at 14nm while others at 22nm, the rest aren't shown which node they use. So it's quite possible that both 14nm & 22nm are used for those parts.Maybe they sold it to someone else, scrapped it or recycled them? The move to 22nm isn't as easy, as many have said here. Intel would need the "tools" & the space for these tools, so if they don't have enough capacity at 22nm then it might be an even bigger problem that what they're letting on.
Though it seems that the list of products moving back might be rather small, they can't backport competing from leading segments where they don't have a monopoly, like networking or NAND.