Intel Officially Reveals What's Coming After Coffee Lake: The 10 nm Ice Lake

[Raevenlord]

A pretty underwhelming post on Intel's official page has pulled the curtains of the company's architecture name post their 8th generation processors. Actually, it's a little more puzzling than that, since Intel is actually detailing the codename of an architecture that's supposed to come right after their 8th generation - read, Coffee Lake - processors. Keep in mind that Coffee Lake, whilst being supposed to bring a reorganization of Intel's product stack in response to AMD's Ryzen success, will still be in the 14 nm++ process - the third such architecture in the same process, after Skylake (14 nm) and Kaby Lake (14 nm+) before it. Cannon Lake, however, is supposed to be the company's first tick into the 10 nm process.

Intel has moved over from their famed tick-tock (where tick is a process shrink and tock is a new architecture on the same process) cadence, and are now telling customers to expect at least three "tocks" per process. It's expected that Intel will launch mobile processors on the 10 nm process before any desktop parts are launched on the same process; this could stem from the fact that mobile parts are typically lower-power, smaller-sized dies, which are easier and cheaper to produce out of a still maturing 10 nm process, which usually implies lower than ideal yields.



View at TechPowerUp Main Site

 

[Steevo]

Intel is getting surpassed on process node as 7nm is already in pre-production and will be out next year by Glo-Flo.

 

[NicklasAPJ]

Intel is getting surpassed on process node as 7nm is already in pre-production and will be out next year by Glo-Flo.

What does it help if it dont perform well.

 

[ppn]

10 nm density is phenomenal. Almost triple that of 14nm say what?
More than 10 fold compared to 32nm Sandy bridge mainstream quad. And what we get is 6-core at best.

 

[efikkan]

Ice Lake will be a new architecture, pushing IPC beyond Skylake.

Intel is getting surpassed on process node as 7nm is already in pre-production and will be out next year by Glo-Flo.

"xx nm" are just marketing terms at this point. In "Intel terms", both TSMC 16nm/12nm FinFet and Samsung 14nm would be classified as "20nm". The new "7nm" node will be more comparable with Intel's current 14nm node. Intel is still >2 years ahead in production technology.

 

[GorbazTheDragon]

"xx nm" are just marketing terms at this point. In "Intel terms", both TSMC 16nm/12nm FinFet and Samsung 14nm would be classified as "20nm". The new "7nm" node will be more comparable with Intel's current 14nm node. Intel is still >2 years ahead in production technology.

I've been wondering about these discrepancies for a while, any idea on articles that take a look at this?

 

[birdie]

This will be the fourth SkyLake reiteration - aside from very minor improvements to the iGPU. I wonder what Intel has been busy with for the past 6 years, since the SkyLake uarch was developed since 2011.

 

[ppn]

What are u talking about. Not only that density went up since 2011's 7.5 to a whooping 100 Mtr/mm2 on 10nm. But they also improved AVX and stuff and IPC got 20% more efficient per clock compared to Sandy. But wonder no further. Adding 2 cores on the mainstream platform should suffice. Isn't it amazing.

 

[webdigo]

So what happens when intel reaches die size 1nm? How can that be improved from that point on?

 

[efikkan]

So what happens when intel reaches die size 1nm? How can that be improved from that point on?

New materials allowing stacked dies. If the energy consumption is low enough and the production cheap enough, we should be able to stack a lot.

 

[brutlern]

I hope Intel realizes that Vanilla Ice was a one hit wonder.

 

[jaggerwild]

What are u talking about. Not only that density went up since 2011's 7.5 to a whooping 100 Mtr/mm2 on 10nm. But they also improved AVX and stuff and IPC got 20% more efficient per clock compared to Sandy. But wonder no further. Adding 2 cores on the mainstream platform should suffice. Isn't it amazing.

You'll need a new motherboard also........

 

[ZoneDymo]

I hope Intel realizes that Vanilla Ice was a one hit wonder.

yeah but Ice Cube is still badass

 

[tehehe]

I've been wondering about these discrepancies for a while, any idea on articles that take a look at this?

http://www.anandtech.com/show/8223/an-introduction-to-semiconductor-physics-technology-and-industry

 

[bubbly1724]

The new "7nm" node will be more comparable with Intel's current 14nm node. Intel is still >2 years ahead in production technology.

 

[TheGuruStud]

Ice Lake will be a new architecture, pushing IPC beyond Skylake.

Don't do drugs kids or you'll think Skylake IV is something new.

 

[PowerPC]

Intel never wants to go beyond 5%-10% each generation. Is that even a surprise by now? All they want is to make money and this anti-consumer strategy already helped them become filthy rich, so why should they change anything? They will always aim for the minimum improvement they can possibly sell through marketing. Even if some scientist at Intel comes up with something that is >10% faster, cooler, better, they'll just put it on the back-burner to sell it to you later. Expecting big leaps from Intel is laughable at this point. They're probably upset they were forced to make a 6-core consumer CPU so early.

 

[goodeedidid]

Intel never wants to go beyond 5%-10% each generation. Is that even a surprise by now? All they want is to make money and this anti-consumer strategy already helped them become filthy rich, so why should they change anything? They will always aim for the minimum improvement they can possibly sell through marketing. Even if some scientist at Intel comes up with something that is >10% faster, cooler, better, they'll just put it on the back-burner to sell it to you later. Expecting big leaps from Intel is laughable at this point. They're probably upset they were forced to make a 6-core consumer CPU so early.

OMG a company in a capitalistic world wants to make profit and money. Isn't this mind blowing.

 

[Hood]

Don't do drugs kids or you'll think Skylake IV is something new.

But we like drugs! I think Intel has been quietly developing an entirely new arch, while just keeping the Core arch viable with little improvements. If a company like AMD, one-tenth the size of Intel, managed to increase IPC 40% going from FX to Ryzen, I think Intel will do the same or better when they're ready to. They're already hinting at such, and Intel doesn't usually brag and then not back it up eventually (yes, Optane is coming along slowly, but it'll get here). So I'm not worried about the next couple of Core updates, my 4th gen does just fine for now. In 2 or 3 years when Intel releases something amazing, then I'll do a major upgrade. It may just be the drugs talking, but I have faith in Intel.

 

[Imsochobo]

What are u talking about. Not only that density went up since 2011's 7.5 to a whooping 100 Mtr/mm2 on 10nm. But they also improved AVX and stuff and IPC got 20% more efficient per clock compared to Sandy. But wonder no further. Adding 2 cores on the mainstream platform should suffice. Isn't it amazing.

avx isn't all it's made up to be, really!.
IPC, give sandy DDR4 and the same clock speeds and that IPC number will be much lower.
Unfortunately we cannot do it, but we can give 7700K 2133 mhz and sandy 1800 mhz and I've seen the numbers, it doesn't look like progress at all!

 

[jigar2speed]

What does it help if it dont perform well.

I wonder how do you know that.

 

[Rictorhell]

I've been very pleased that AMD is back in the picture, for something more than just graphics cards. I like that when it comes to technology and semi-conductor news that there is now someone being mentioned on a fairly regular basis, other than Intel.

Intel has been slacking off quite a bit for the past few years and they haven't really had a worthy competitor for a while now.

My next computer probably will be built around an AMD Threadripper CPU and I am very grateful to AMD for taking steps to get their act together and give Intel at least some amount of competition. I hope AMD continues to compete with Intel and grow more and more.

 

[Kronauer]

Looks like Intel is really trying to build a hype-train, so people start to talk about Intel chips, instead of AMD.

But this "will make 10nm chips, i promise" is outdated. According to Intel we should already have 10nm chips.
But Intel itself screwed it up.

Cuz instead of 14nm shrink in 2014 we got haswell refresh. In 2015 instead of 14nm new architecture we got broadwell (the 14nm shrink of haswell). In 2016 instead of 10nm shrink, we got 14nm architecture a.k.a. Skylake and in 2017 instead of 10nm new architecture we got "skylake refresh"; and if Ryzen wouldnt have been this successful, in 2018 we would've got just another refresh of kaby lake.

But right now they are announcing 2 generations a CPU families? Even though Intel still holds the vast majority of the x86 CPU market, this looks like a very desperate cry for attention.

 

[efikkan]

It's sad to see that any article about Intel these days will be met with pages of bashing in the forums.

avx isn't all it's made up to be, really!

AVX gives tremendous performance gains, in the range of ~10-50×. But the program has to be designed to take advantage of it, and the workload has to be a fairly large continuous stream of data. Programmers use intrinsics to utilize AVX in their software, which are just macros which are almost directly mapped to assembly. AVX support is not something that can be put in last minute in an application, it has to be designed to take advantage of it.

 

[R0H1T]

So what happens when intel reaches die size 1nm? How can that be improved from that point on?

Well you can;t go to 1nm, Si node shrinks should stop at 4/5 nm (depends on Fab's marketing numbers) & beyond that there's tunnelling to contend with.