Intel Coffee Lake and Cannonlake in H2-2017, Icelake in 2018 (DigiTimes)

[jpiniero]

Another thought... You know, Intel did say they were cutting back on PC R&D. Would it really shock you if Cannonlake Xeon D was the first product out on 10 nm?

 

[Sweepr]

Would be fun if the first 10nm product was Gemini Lake, but most people I talked to seem think it's still 14nm.

 

[Arachnotronic]

If you look this is a roadmap from April.

Hint: Intel roadmaps always change in one way...products get delayed.

 

[Arachnotronic]

Another thought... You know, Intel did say they were cutting back on PC R&D. Would it really shock you if Cannonlake Xeon D was the first product out on 10 nm?

The R&D isn't intended to impact CPU/SoC development, but more platform development, software, etc.

 

[dark zero]

AMD and VIA won some time to perfect their next gen cores.

 

[Arachnotronic]

AMD and VIA won some time to perfect their next gen cores.

Not really, no.

 

[Arachnotronic]

Would be fun if the first 10nm product was Gemini Lake, but most people I talked to seem think it's still 14nm.

Yep, Gemini is 14nm.

 

[dooon]

The Motley Fool
Exclusive The Secret of Intel Corporation's Coffee Lake Revealed
http://www.fool.com/investing/2016/09/14/exclusive-the-secret-of-intel-corporations-coffee.aspx

In trying to find more information out about Coffee Lake, I stumbled across the following on Intel's website:

Cannon Lake H (Coffee Lake)
Cannon Lake S (Coffee Lake)
Cannon Lake U/Y

 

[witeken]

Many of his articles are quite clickbait, but sometimes he does have meaningful things to say.

 

[mikk]

If it comes from an Intel website, then it pretty much confirms that Coffe Lake is Cannonlake in 14nm. This is a good news. With this in mind it makes perfect sense that after Coffe Lake Intel goes to Icelake for desktop.

 

[Arachnotronic]

If it comes from an Intel website, then it pretty much confirms that Coffe Lake is Cannonlake in 14nm. This is a good news. With this in mind it makes perfect sense that after Coffe Lake Intel goes to Icelake for desktop.

yeah, I was worried that Intel would stick with the same SKL architecture for three generations -- that would have been sheer laziness. CNL probably won't bring any big bangs, but perf/clock should go up, core count goes up, and things generally get a lot better.

 

[Azuma Hazuki]

So from the look of it, they more or less have the Cannonlake uarch finalized, but they're going to do performance parts on a well-vetted 14nm process, and drop to 10nm for the Y-series due to...power saving?

 

[jpiniero]

yeah, I was worried that Intel would stick with the same SKL architecture for three generations -- that would have been sheer laziness. CNL probably won't bring any big bangs, but perf/clock should go up, core count goes up, and things generally get a lot better.

Backporting Cannon Lake to 14nm+ would be pretty expensive though, wouldn't you think?

TBH if this true I think this means that there aren't much in the way of changes between CNL/CFL and KBL. So I wouldn't expect any IPC increase at all, and AVX-512 is likely to be emulation only.

I had always pegged Canon Lake as to when Intel was going to release a "Core Phone" processor. Even though that's dead, power consumption above everything else was still likely their #1 priority in development.

 

[Arachnotronic]

Backporting Cannon Lake to 14nm+ would be pretty expensive though, wouldn't you think?

Not as expensive as missing out on a product cycle (aka trying to fight Zen APUs with Kaby Lake for a whole year) or taking a large hit on margins because 10nm yields are terrible. Intel's R&D scale is gigantic, doing a Cannon Lake chip on 14nm is hardly the most expensive thing for them. Come on now.

TBH if this true I think this means that there aren't much in the way of changes between CNL/CFL and KBL. So I wouldn't expect any IPC increase at all, and AVX-512 is likely to be emulation only.

CNL does bring core/IPC improvements over Skylake/Kaby Lake.

I had always pegged Canon Lake as to when Intel was going to release a "Core Phone" processor. Even though that's dead, power consumption above everything else was still likely their #1 priority in development.

TDP for CNL-Y is 5.2W, so expect perf to go up within a similar power envelope to KBL.

 

[nvgpu]

http://www.fool.com/investing/2016/09/14/exclusive-the-secret-of-intel-corporations-coffee.aspx

Coffee Lake is Cannon Lake but on Intel 14nm process?

 

[witeken]

People are very pessimistic about 10nm. I'm quite excited for it. From now on we will see a process change every year, how cool is that? Maybe 10nm w/ FinFET with Cannonlake, but then with 10nm+ they might add III-V and Ge, then with 10nm++ add quantum well FETs.

Okay, that prolly won't happen, but it's good to see them being more agile and still have a solid pipeline of 1 product per year. You can't really ask for all the much more. Of course we want big IPC bumps and IGP increases, but we don't have any infos about that.

 

[Arachnotronic]

http://www.fool.com/investing/2016/09/14/exclusive-the-secret-of-intel-corporations-coffee.aspx

Coffee Lake is Cannon Lake but on Intel 14nm process?

Correct

 

[Arachnotronic]

People are very pessimistic about 10nm.

For good reason. Coffee Lake tells you that 10nm yields are still in the crapper and are going to be until pretty much the end of 2018. They can get away with a 14nm Coffee Lake because the segments it will sell into are not really battery/power sensitive (not at all for desktops, minimally for big bulky laptops that people usually keep plugged in anyway).

I'm quite excited for it. From now on we will see a process change every year, how cool is that? Maybe 10nm w/ FinFET with Cannonlake, but then with 10nm+ they might add III-V and Ge, then with 10nm++ add quantum well FETs.

It'll be interesting to see what they do. They probably won't make dramatic changes to gate pitch over that time so density will be mostly unaffected, but they can play with different materials, fin heights, etc. Basically, this is a way for them to bring out process improvements while they figure out how to shrink. Shrinking is quite hard especially without EUV on the litho side.

Okay, that prolly won't happen, but it's good to see them being more agile and still have a solid pipeline of 1 product per year. You can't really ask for all the much more. Of course we want big IPC bumps and IGP increases, but we don't have any infos about that.

Tigerlake remains a mystery. I hope they do CPU core improvements & use Gen12 GPU to really get the most out of that shiny 10nm++ process.

 

[jpiniero]

TDP for CNL-Y is 5.2W, so expect perf to go up within a similar power envelope to KBL.

Probably related to the FIVR and higher base clocks.

 

[Maxima1]

They can get away with a 14nm Coffee Lake because the segments it will sell into are not really battery/power sensitive (not at all for desktops, minimally for big bulky laptops that people usually keep plugged in anyway).

Not all laptops are bulky and big like an Alienware. There are thinner models that should have Nvidia 1050 (e.g. Dell XPS or 7000 series) with standard voltage CPUs. Those generally have twice the battery capacity.

 

[ShintaiDK]

AVX-512 is likely to be emulation only.

What performance do you think emulation would be. Ye, that's right. Its not going to happen.

 

[witeken]

For good reason. Coffee Lake tells you that 10nm yields are still in the crapper and are going to be until pretty much the end of 2018. They can get away with a 14nm Coffee Lake because the segments it will sell into are not really battery/power sensitive (not at all for desktops, minimally for big bulky laptops that people usually keep plugged in anyway).

Do keep in mind that the rules have changed by being on a node for 3 years. The yield of 14nm hasn't even been *healthy* for all that long, but they will already transition to the immature 10nm next year. Releasing products on 14nm isn't as drmatic since they will still have a few years (1 more year) to transition to 10nm. Sad but true.

 

[Nothingness]

What performance do you think emulation would be. Ye, that's right. Its not going to happen.

I guess he meant mapping AVX-512 onto (enhanced) AVX2 units. Or at least that's how I understood it

 

[Sweepr]

Correct

Nice find there! So Coffee Lake is way more than Skylake/Kaby Lake on steroids. We're getting the complete package (CPU + iGPU architecture uodate) on top of 'more cores' (6C/12T).

 

[jpiniero]

What performance do you think emulation would be. Ye, that's right. Its not going to happen.

You know what I mean. The full hardware needed to do AVX-512 properly would be a waste on the mainstream CPU given the IGP.