Discussion Intel current and future Lakes & Rapids thread

[jpiniero]

Where does the chrisdar info say 2019?
It appears to have been posted 6 months ago, and nothing else since?

What am I missing?

IIRC Intel said recently that they moved the 10 nm HVM from the middle of this year to the endish. That was probably more due to Cannonlake's demise but it also kind of confirms that Icelake U/Y won't be available until next year.

 

[Dayman1225]

IIRC Intel said recently that they moved the 10 nm HVM from the middle of this year to the endish. That was probably more due to Cannonlake's demise but it also kind of confirms that Icelake U/Y won't be available until next year.

Intel has stuck with the small ramp up in H1 with a bigger ramp in H2 AFAIK, no changes that I am aware of.

 

[jpiniero]

Intel has stuck with the small ramp up in H1 with a bigger ramp in H2 AFAIK, no changes that I am aware of.

In order to launch Icelake U/Y by Q4, you have to start HVM at the beginning of H2 and not the end. I think it was subtle in terms of the change. Still seems like it should be a solid product provided the turbo clocks are competitive (3.6 for Y and 4.2 for U).

 

[Eug]

This 8th gen is really an interesting mess, with implications for 9th gen.

Kaby Lake Refresh and Coffee Lake are 8th gen 14 nm, but both have additional cores. The lone leak for a Cannon Lake Y 10 nm part is still dual-core. Yeah I realize that at least superficially, Cannon Lake was originally supposed to be a plain die shrink but still...

So, does this mean that Intel has partially scrapped the "Process" stage, and instead will effectively be combining "Process" and "Architecture" in the next release with Ice Lake?

Also, are we expecting 15 W U parts with Iris Plus any time soon? Would that be Coffee Lake?

 

[IntelUser2000]

Also, are we expecting 15 W U parts with Iris Plus any time soon? Would that be Coffee Lake?

Yea, its called Coffeelake-U and is launching alongside Coffeelake-H parts, you know the really high clocked 6 core Coffeelake mobile chips like the 8950HK. Probably around early April like mainstream desktop Coffee Lake.

Coffeelake-U is 4+3e.

 

[Dayman1225]

Yea, its called Coffeelake-U and is launching alongside Coffeelake-H parts, you know the really high clocked 6 core Coffeelake mobile chips like the 8950HK. Probably around early April like mainstream desktop Coffee Lake.

Coffeelake-U is 4+3e.

Talking about CFL U, it just appeared on GFXBench, i5 8269U with Iris Plus Graphics 650

 

[Bouowmx]

Why was 4+3e selected to be ported to 14 nm ++, rather than 4+2?

 

[jpiniero]

Why was 4+3e selected to be ported to 14 nm ++, rather than 4+2?

Kaby-R came out much earlier and is part of this generation. I think the CFL 4+3e is 28W only (7th gen had both 15W and 28W), but we'll see.

 

[Dayman1225]

Kaby-R came out much earlier and is part of this generation. I think the CFL 4+3e is 28W only (7th gen had both 15W and 28W), but we'll see.

Old roadmaps say that Intel has planned for 15w parts but as this is quite old it may have changed for CFL

 

[Eug]

Yea, its called Coffeelake-U and is launching alongside Coffeelake-H parts, you know the really high clocked 6 core Coffeelake mobile chips like the 8950HK. Probably around early April like mainstream desktop Coffee Lake.

Coffeelake-U is 4+3e.

Kaby-R came out much earlier and is part of this generation. I think the CFL 4+3e is 28W only (7th gen had both 15W and 28W), but we'll see.

So which is correct? I was asking about 15 W quad-core GT3e.

 

[jpiniero]

So which is correct? I was asking about 15 W quad-core GT3e.

You'll have to wait and find out. Maybe it depends on what Apple wants.

 

[Arachnotronic]

You'll have to wait and find out. Maybe it depends on what Apple wants.

There will be 15W GT3e CFL parts.

 

[IntelUser2000]

So which is correct? I was asking about 15 W quad-core GT3e.

That roadmap is still valid aside from Cannonlake, which was affected by process delays.

 

[Eug]

You'll have to wait and find out. Maybe it depends on what Apple wants.

If it is dependent upon Apple, then it will be 15 W GT3e, cuz that is what they use now.

BTW, the current i7-8650U quad GT2 is 1.9 GHz. Would we expect quad GT3e to be lower clocked? Say 1.6-1.7 GHz?

That roadmap is still valid aside from Cannonlake, which was affected by process delays.

Hope so.

 

[Arachnotronic]

If it is dependent upon Apple, then it will be 15 W GT3e, cuz that is what they use now.

BTW, the current i7-8650U quad GT2 is 1.9 GHz. Would we expect quad GT3e to be lower clocked? Say 1.6-1.7 GHz?


Hope so.

Quad GT3e will clock higher because of more efficient process, improved physical implementation, more aggressive binning, and more efficient PCH (14nm vs 22nm, leaving more power budget for the actual compute part of the chip).

 

[IntelUser2000]

If it is dependent upon Apple, then it will be 15 W GT3e, cuz that is what they use now.

BTW, the current i7-8650U quad GT2 is 1.9 GHz. Would we expect quad GT3e to be lower clocked? Say 1.6-1.7 GHz?

Hope so.

Base clocks are lower on GT3e, because it has to power more GPU, and base clocks are more or less worst case. CPU only performance isn't affected though, because GPU is power gated when not used.

Don't see why the roadmap needs changing. Coffeelake isn't affected by process delays.

 

[Eug]

Quad GT3e will clock higher because of more efficient process, improved physical implementation, more aggressive binning, and more efficient PCH (14nm vs 22nm, leaving more power budget for the actual compute part of the chip).

Hmmm... That seems rather optimistic.

Anyhoo, I will go on record here to arbitrarily predict that 15 W Coffee Lake U quad-core with GT3e will be ~1.7 GHz.

 

[Eug]

So, anyone believe this fanlesstech guy’s twitter post from 2017-12?

https://mobile.twitter.com/FanlessTech/status/941020311314673665

CNL-Y is still on track for June 2018.

 

[imported_bman]

So, anyone believe this fanlesstech guy’s twitter post from 2017-12?

https://mobile.twitter.com/FanlessTech/status/941020311314673665

CNL-Y is still on track for June 2018.

Makes sense, in time for WWDC 2018 for an upgrade of the Macbook Retina from Kaby-Y to Cannon-Y. Though if the rumors about Icelake-Y are true then it is a skip it part for anyone other than people who can't wait till 2019.

 

[IntelUser2000]

CNL-Y is still on track for June 2018.

Yea, they are quite reliable.

CNL-Y series in Mid-2018
Whiskey Lake U for BTS 2018
Coffeelake H/U for April/May 2018
Coffeelake mainstream for April 2018
Coffeelake 8 core for BTS 2018
Cascade Lake SP for Q3 2018
Cascade Lake X for Q4 2018

So not a single Icelake will be seen until at least spring of 2019.

 

[Eug]

Yea, they are quite reliable.

CNL-Y series in Mid-2018
Whiskey Lake U for BTS 2018
Coffeelake H/U for April/May 2018
Coffeelake mainstream for April 2018
Coffeelake 8 core for BTS 2018
Cascade Lake SP for Q3 2018
Cascade Lake X for Q4 2018

So not a single Icelake will be seen until at least spring of 2019.

So perhaps my earlier prediction is right after all.

2017: Kaby Lake Y 4.5 Watt dual-core (1.0 GHz base clock for 7Y30). No Thunderbolt.
2018: Cannon Lake Y 5.2 Watt dual-core (1.1 GHz base clock for entry level). No Thunderbolt.
2019: Ice Lake Y 5.2 Watt quad-core (no clockspeeds known yet). Includes Thunderbolt 3.

 

[jpiniero]

So, anyone believe this fanlesstech guy’s twitter post from 2017-12?

https://mobile.twitter.com/FanlessTech/status/941020311314673665

CNL-Y is still on track for June 2018.

It may have been... but not now.

That's why I think there's those new rumors about Apple making a new 'cheap' Macbook Air are true. It's something of a temporary replacement for the Macbook. Maybe it'll come with the Kaby Lake 8130U which has already been announced and looks like the Cannonlake U replacement.

 

[Arachnotronic]

Base clocks are lower on GT3e, because it has to power more GPU, and base clocks are more or less worst case. CPU only performance isn't affected though, because GPU is power gated when not used.

Don't see why the roadmap needs changing. Coffeelake isn't affected by process delays.

On Kaby Lake, base clocks were slightly lower for GT3e configuration vs GT2 configuration (though turbo was higher), but that was apples-to-apples in terms of process/physical layout.

14nm++ provides a reasonable efficiency boost and the migration to 14nm SoC for the PCH from 22nm SoC should also help there. I wouldn't be surprised to see the GT3e models match/exceed the Kaby Lake-R models in base clock (though Whiskey Lake on 14nm++ and with the new PCH would be a different matter).

 

[Dayman1225]

Icelake and 10nm+ are on track - FanlessTech

Officially launching before the end of the year (U and Y series) but from what we heard, Ice Lake is a 2019 product. - FanlessTech

 

[IntelUser2000]

So perhaps my earlier prediction is right after all.

The TB support could mean in the same way as with the 300-series chipset: https://hothardware.com/gallery/NewsItem/41741?image=big_coffee_lake_features.jpg&tag=popup

It's not integrated but "supported".

14nm++ provides a reasonable efficiency boost and the migration to 14nm SoC for the PCH from 22nm SoC should also help there.

Chipset TDPs haven't really gone down. The die sizes and power consumption are kept the same level because they add features. 300-series chipsets have improved audio DSP, integrated SD controller, WiFi MAC, and updated USB. Integration might reduce overall platform power, both idle and load but that just means the burden goes to the chipset instead.

The question of whether we'll see CFL-U at similar clocks to KBL-R is really an academic matter. Because vast majority of the times that makes effective CPU clocks lower will be in gaming, and GT3e parts are going to be significantly faster anyway. When the CPU is active, there will be no frequency loss.