Intel Skylake / Kaby Lake

[LTC8K6]

I still don't know how will they reach 95W TDP at 4,3GHz with 6 core. I know "TDP" is not for power consumption, but lately everybody is basically using it as power consumption.
Maybe with base clock 3,7GHz, but even that could be hard to reach.

14nm++ is apparently noticeably more efficient. Intel said it was better than their initial 10nm process.

 

[NTMBK]

14nm++ is apparently noticeably more efficient. Intel said it was better than their initial 10nm process.

That might be more of a comment on the state of Intel's 10nm...

 

[AtenRa]

I thought Coffeelake was 14nm+ not ++

 

[LTC8K6]

I thought Coffeelake was 14nm+ not ++

No, it's ++, which is part of the reason for the all the hopes.

 

[Lodix]

Per Intel's slides 14nm++ brings around the same performance/efficiency jump over 14nm+, that it did to the first 14nm.

 

[Arachnotronic]

That might be more of a comment on the state of Intel's 10nm...

Intel did a solid job improving the performance of its 14nm variants -- 14nm+ was clearly nice, and 14nm++ looks like another winner in terms of performance.

It's a shame that these were added somewhat at the last minute; had Intel planned for this years out, they could've done new architectures co-optimized with the enhanced processes, which could've made for some even more impressive products.

I think for 10nm, Intel had no delusions of a 2 year cycle to 7nm, though, so hopefully that will mean good things for Tiger Lake and maybe the next part after that, if it is also on 10nm.

 

[dullard]

I still don't know how will they reach 95W TDP at 4,3GHz with 6 core. I know "TDP" is not for power consumption, but lately everybody is basically using it as power consumption.
Maybe with base clock 3,7GHz, but even that could be hard to reach.

It is all clear to me from this chart (which was part of an Intel presentation this spring).
https://en.wikichip.org/wiki/File:intel_14nm++.png

The red curve is Skylake that the Coffee Lake is basically based upon. The orange curve is Coffee Lake. The average power loss per fin is about 52% lower on Coffee Lake compared to Skylake (assuming you are going for the same performance).

Now for some very crude math (yes the math could be done better).

• Skylake 6700K: 91 W / 4.0 GHz max turbo / 4 max cores = 5.688 W/GHz/Core
• Coffee Lake 8700K rumor: 95 W / 4.3 max turbo / 6 max cores = 3.682 W/GHz/Core

Now what is the difference in power between those two numbers? 5.688 / 3.682 = 1.54. The Skylake 6700K uses 54% more power per max clock, which matches up very closely with the 52% power savings of the 14nm++ process.

Of course that math was with TDP and not actual power used, and it isn't take into account many fine details (more or less thermal throttling, a chip doesn't have to stay on the same performance point on the curve and can get more or less power that way, software capability of keeping 12 threads active, etc). But it shows you roughly how it can be done.

 

[IRobot23]

It is all clear to me from this chart (which was part of an Intel presentation this spring).
https://en.wikichip.org/wiki/File:intel_14nm++.png

The red curve is Skylake that the Coffee Lake is basically based upon. The orange curve is Coffee Lake. The average power loss per fin is about 52% lower on Coffee Lake compared to Skylake (assuming you are going for the same performance).

Now for some very crude math (yes the math could be done better).

• Skylake 6700K: 91 W / 4.0 GHz max turbo / 4 max cores = 5.688 W/GHz/Core
• Coffee Lake 8700K rumor: 95 W / 4.3 max turbo / 6 max cores = 3.682 W/GHz/Core

Now what is the difference in power between those two numbers? 5.688 / 3.682 = 1.54. The Skylake 6700K uses 54% more power per max clock, which matches up very closely with the 52% power savings of the 14nm++ process.

Of course that math was with TDP and not actual power used, and it isn't take into account many fine details (more or less thermal throttling, a chip doesn't have to stay on the same performance point on the curve and can get more or less power that way, software capability of keeping 12 threads active, etc). But it shows you roughly how it can be done.

It does not work like that, we can see that i7 7700K is prety power hungry, i7 7820K is power hungry.... if you want efficiency, you probably won't reach high clocks. You just cannot have both, low power or high performance and you have to choose.

Intel 14nm++... I think its only marketing, probably little optimization to get higher clocks.

 

[Arachnotronic]

Intel 14nm++... I think its only marketing, probably little optimization to get higher clocks.

It's not.

 

[dullard]

It does not work like that, we can see that i7 7700K is prety power hungry, i7 7820K is power hungry.... if you want efficiency, you probably won't reach high clocks. You just cannot have both, low power or high performance and you have to choose.

Intel 14nm++... I think its only marketing, probably little optimization to get higher clocks.

Yes, you get to choose. But you get to choose anywhere on a given power curve such as in the graph that I showed. You can choose something that is more performance and more power on the red curve (upper right corner), or you can choose something that is less performance and less power on the red curve (lower left corner).

When you have a new process, you have a new curve to choose from (such as the orange curve instead of the red curve). In this case, Intel's new 14 nm++ power curve is far superior to their original 14 nm power curve. So superior that Intel is going that route and not 10 nm for now (yield issues also play a role in that decision).

 

[LTC8K6]

It does not work like that, we can see that i7 7700K is prety power hungry, i7 7820K is power hungry.... if you want efficiency, you probably won't reach high clocks. You just cannot have both, low power or high performance and you have to choose.

Intel 14nm++... I think its only marketing, probably little optimization to get higher clocks.

Well, it's def not "only" marketing. 14++ is clearly an improved process, and there's no reason to think it wouldn't result in a better chip.

 

[IRobot23]

It's not.

Please do explain, it is only my thinking.. why would they waste their money on optimizing 14nm?

 

[LTC8K6]

Please do explain, it is only my thinking.. why would they waste their money on optimizing 14nm?

Because 10nm proved quite difficult. That was well known.
Intel has been pretty clear that 14++ is better than their 10nm, even though they demoed a 10nm laptop quite a while ago.

 

[IRobot23]

Because 10nm proved quite difficult. That was well known.
Intel has been pretty clear that 14++ is better than their 10nm, even though they demoed a 10nm laptop quite a while ago.

Fair point, still hard to believe those 4,3GHz x 6 Cores (would be amazing) ... we will see, I guess.

 

[LTC8K6]

Fair point, still hard to believe those 4,3GHz x 6 Cores (would be amazing) ... we will see, I guess.

I have been suspicious of the 8700K clocks as well. I still suspect they won't be quite that good even though we seem to have solid info now.

 

[swilli89]

So is this a new architecture ala "Tock"? Akin to Ivy Bridge to Haswell or Broadwell to Skylake? Or is this basically 6-core Skylake on a more mature 14nm?

I know the Intel cycle has been changed to Process - Architecture - Optimization.. so if Broadwell was the 14nm move and Skylake was the new architecture.. that means Kaby Lake was the "optimization".

Is Coffee Lake a second optimization now? Used to be so easy to keep up with Intel's new products but its getting more difficult.

 

[Dayman1225]

As far as we know its Skylake on 14nm++ with more cores. Though there was rumours that it was backported Cannonlake, the next Arch change should be Icelake, though I suppose that could be counted as process too, for desktop.

 

[RichUK]

It is all clear to me from this chart (which was part of an Intel presentation this spring).
https://en.wikichip.org/wiki/File:intel_14nm++.png

The red curve is Skylake that the Coffee Lake is basically based upon. The orange curve is Coffee Lake. The average power loss per fin is about 52% lower on Coffee Lake compared to Skylake (assuming you are going for the same performance).

Now for some very crude math (yes the math could be done better).

• Skylake 6700K: 91 W / 4.0 GHz max turbo / 4 max cores = 5.688 W/GHz/Core
• Coffee Lake 8700K rumor: 95 W / 4.3 max turbo / 6 max cores = 3.682 W/GHz/Core

Now what is the difference in power between those two numbers? 5.688 / 3.682 = 1.54. The Skylake 6700K uses 54% more power per max clock, which matches up very closely with the 52% power savings of the 14nm++ process.

Of course that math was with TDP and not actual power used, and it isn't take into account many fine details (more or less thermal throttling, a chip doesn't have to stay on the same performance point on the curve and can get more or less power that way, software capability of keeping 12 threads active, etc). But it shows you roughly how it can be done.

TDP is a thermal measure, not power.

 

[Markfw]

It's not.

And you know because ? You work for Intel ?

 

[LTC8K6]

TDP is a thermal measure, not power.

Well, here is what Intel says it is:

TDP
Thermal Design Power (TDP) represents the average power, in watts, the processor dissipates when operating at Base Frequency with all cores active under an Intel-defined, high-complexity workload. Refer to Datasheet for thermal solution requirements.

A power measure.

 

[mikk]

Before Kabylake people told 14nm+ is only marketing and it wasn't, it really pushed the clock wall 200-300 Mhz higher. Same story with 14nm++ now. Even KBL-R on 14nm+ received some enhancements over the original 14nm+. Some people should open their eyes and pay attention to the Turbo speeds of Coffeelake.

 

[PeterScott]

Well, here is what Intel says it is:
A power measure.

This seems like something of a semantics argument. Watts can represent input power, it can also represent a thermal load that needs to be dissipated. Every input power watt, is going to to turn into a thermal load that needs to be dissipated, so they are essentially interchangeable. Watts in (electricity) == Watts out (thermal load).

But since Intel calls it "Thermal" Design Power, and talks about: "...watts, the processor dissipates...", it does seem to be more about the heat dissipation side of things.

 

[raghu78]

Before Kabylake people told 14nm+ is only marketing and it wasn't, it really pushed the clock wall 200-300 Mhz higher. Same story with 14nm++ now. Even KBL-R on 14nm+ received some enhancements over the original 14nm+. Some people should open their eyes and pay attention to the Turbo speeds of Coffeelake.

fyi clock speed increases happened in the past too. Its just that Intel did not market it as they are doing now because their process roadmap has slowed down badly at 10nm.

https://semiaccurate.com/2017/04/04/intels-hyperscaling-is/

http://www.anandtech.com/show/7003/the-haswell-review-intel-core-i74770k-i54560k-tested
http://www.anandtech.com/show/8227/devils-canyon-review-intel-core-i7-4790k-and-i5-4690k

4770k - 4790k brought 500 Mhz increase in base and turbo clocks. For turbo clocks 6700k -> 7700k brought 300 Mhz.and 7700k - 8700k another 200 Mhz. Coffeelake brings 50% more cores over the previous gen Skylake/Kabylake. Thats the major improvement and what is different from the past generations. But then its more in response to the change in competitive dynamics in the PC industry.

 

[Arachnotronic]

fyi clock speed increases happened in the past too. Its just that Intel did not market it as they are doing now because their process roadmap has slowed down badly at 10nm.

https://semiaccurate.com/2017/04/04/intels-hyperscaling-is/

http://www.anandtech.com/show/7003/the-haswell-review-intel-core-i74770k-i54560k-tested
http://www.anandtech.com/show/8227/devils-canyon-review-intel-core-i7-4790k-and-i5-4690k

4770k - 4790k brought 500 Mhz increase in base and turbo clocks. For turbo clocks 6700k -> 7700k brought 300 Mhz.and 7700k - 8700k another 200 Mhz. Coffeelake brings 50% more cores over the previous gen Skylake/Kabylake. Thats the major improvement and what is different from the past generations. But then its more in response to the change in competitive dynamics in the PC industry.

This isn't the same thing. 4790K was basically an overclocked 4770K -- if you look at the reviews and user comments, the maximum frequency ceiling for the 4790K was really no better than the 4770K's.

The 7700K had a clear clock ceiling improvement of between 200MHz and 300MHz as mikk said, and there were also pretty large improvements in sustained frequency/peak frequency on the low power mobile parts, which we didn't really see with the Haswell Refresh.

Since Intel actually made some real changes with 14nm+ compared to 14nm that yielded measurable improvements, there's no reason to not take Intel at its word when it says it's making another significant jump in transistor performance with 14nm++.

 

[mikk]

4770k - 4790k brought 500 Mhz increase in base and turbo clocks. For turbo clocks 6700k -> 7700k brought 300 Mhz.and 7700k - 8700k another 200 Mhz.

That's exactly my point. People told it's only marketing and as you can see they were wrong.