Intel's 14nm Process Smaller Pitch, Smaller Die, Same Power Consumption

[biostud]

So Broadwell-E should be good?

 

[Dresdenboy]

Fellow forum member Dufus kindly provided me with the voltage Data about his Notebook BW core i7-5500U after my request. According to him the VIDs were read from CPU registers using some kind of programming magic. His values are compared to the Speedstep Auto voltages of my G3258 which were read from CPU-Z forcing a power state with windows power settings. I used only data below base frequency to keep things straight and simple.

The position of the Voltages on the Y-Axis isn't really saying much, because here we compare a cheapo Pentium with a ULV-mobile-CPU. What matters is the slope or angle for the Voltage - Frequency - Curve.
IMHO this confirms and demonstrates that Broadwell uses a 14nm low-power-process, at least for this 15 W CPU.
The BW-straight has a slope that is 26% steeper than than of HW, meaning that ascending multipliers require a voltage bump that is by x1.26 higher than those of HW.
I chose red and green based on this graph posted earlier.

The BDW core is a bit smaller than SKL and also looks like having a somewhat altered floorplan. So the crossover with HSW might not just be due to a different fabbing process, but also caused by the design. This is a bit like with XV here:

Frequency is on the Y axis and power is on the X axis.

If we'd have the voltages instead of frequency, the curves should also be more linear. But the crossover point will still be there.

 

[IntelUser2000]

Atom uses a different process variant, the SoC process. That may well be what borked the performance, it is optimized for density instead of performance.

Do you think the engineers did not know that when designing Airmont? Low power processes, which are supposedly "less performing" than Intel's low power process, certainly does not hinder ARM competitors from trouncing Atom chips.

It's up for the engineer to know the limitations of what they are working on and make the best of it. Of course I am not lessening the impact of management/chief architect either.

Also, its been since 32nm since they claimed they were using SoC process for Atom! Something screwed up massively at 14nm. Decades of failed promises are finally converging on at Intel.

 

[NTMBK]

Do you think the engineers did not know that when designing Airmont? Low power processes, which are supposedly "less performing" than Intel's low power process, certainly does not hinder ARM competitors from trouncing Atom chips.

It's up for the engineer to know the limitations of what they are working on and make the best of it. Of course I am not lessening the impact of management/chief architect either.

Also, its been since 32nm since they claimed they were using SoC process for Atom! Something screwed up massively at 14nm. Decades of failed promises are finally converging on at Intel.

Depends on what your definition of success is. Cherry Trail is a tiny die which still crams in a bunch of useful-for-tablets features which Bay Trail lacked: http://www.fool.com/investing/gener...rporation-cherry-trail-die-size-revealed.aspx CPU performance was evidently pretty far down the list. Basically seems to be "port it to 14nm and don't screw it up too much, then move onto the other things".

It's the lack of things like integrated imaging sensors which caused Bay Trail to have such a high "platform cost" compared to its ARM rivals- without an integrated sensor, you need to put a discrete image sensor on the motherboard. This is one of the causes of the massively costly contra-revenue program. Intel's main priority with Cherry Trail was to reduce the massive losses from mobile.

 

[Arachnotronic]

Depends on what your definition of success is. Cherry Trail is a tiny die which still crams in a bunch of useful-for-tablets features which Bay Trail lacked: http://www.fool.com/investing/gener...rporation-cherry-trail-die-size-revealed.aspx CPU performance was evidently pretty far down the list. Basically seems to be "port it to 14nm and don't screw it up too much, then move onto the other things".

It's the lack of things like integrated imaging sensors which caused Bay Trail to have such a high "platform cost" compared to its ARM rivals- without an integrated sensor, you need to put a discrete image sensor on the motherboard. This is one of the causes of the massively costly contra-revenue program. Intel's main priority with Cherry Trail was to reduce the massive losses from mobile.

Actually, Bay Trail had an integrated ISP and all of the things (sans modem) that you'd need in a modern SoC aside from connectivity.

Bay Trail's problem was that the motherboard designs that Intel provided around the SoC were very expensive. They had to do a lot of work with Bay Trail-CR to reduce the platform/board costs for low-cost tablets, and even that work wasn't enough to eliminate contra-revenue (but it took it from something along the lines of $20/unit to maybe $10/unit avg).

Cherry Trail probably brought that down further, but IIRC it was Broxton & SoFIA that were supposed to end contra-revenue in their respective market segments.

 

[Arachnotronic]

It's up for the engineer to know the limitations of what they are working on and make the best of it. Of course I am not lessening the impact of management/chief architect either.

The Airmont chief architect now works at Apple, so clearly he's strong. I think Intel just forced him to fight with one arm tied behind his back. I doubt that will be the case in Cupertino.

 

[NTMBK]

Actually, Bay Trail had an integrated ISP and all of the things (sans modem) that you'd need in a modern SoC aside from connectivity.

Bay Trail's problem was that the motherboard designs that Intel provided around the SoC were very expensive. They had to do a lot of work with Bay Trail-CR to reduce the platform/board costs for low-cost tablets, and even that work wasn't enough to eliminate contra-revenue (but it took it from something along the lines of $20/unit to maybe $10/unit avg).

Cherry Trail probably brought that down further, but IIRC it was Broxton & SoFIA that were supposed to end contra-revenue in their respective market segments.

Gah, got confused with Skylake vs Broadwell. Disregard what I wrote :thumbsup:

 

[el etro]

So Broadwell-E should be good?

Even if can't clock higher than Haswell-E at same TDP/core numbers, it will have more cores.

This according to what we are seeing now.


I think if you don't care much about heavy MT loads you should get a Haswell-E.

 

[poofyhairguy]

I think Intel just forced him to fight with one arm tied behind his back.

What do you mean by that? From my understanding the sacrifices Intel made for Atom's IPC were mostly related to power consumption. Do you believe the (what I thought was a crackpot) theory that Intel didn't want Atom to be too good because that would threaten Core?

 

[Arachnotronic]

What do you mean by that? From my understanding the sacrifices Intel made for Atom's IPC were mostly related to power consumption. Do you believe the (what I thought was a crackpot) theory that Intel didn't want Atom to be too good because that would threaten Core?

It's not crackpot, it's real. Intel can't afford Atom to be anywhere close to Core performance (they've said that Core needs to be 2x Atom in the same power envelope, IIRC), and they need the die sizes to be very small/cheap.

At the same time, Atom is not the top priority for Intel; it's not going to get the resources that Core is going to get.

For example, Intel has two big Core teams -- one in Oregon, one in Haifa, this is how they do the "tick-tock." For Atom, they have just one team in Austin, TX.

This limits the scope of what the company can ultimately achieve with Atom, impacting its competitiveness in the marketplace.

The Atom team had/has talented people within its ranks, but they are not given a real chance to succeed.

 

[poofyhairguy]

It's not crackpot, it's real. Intel can't afford Atom to be anywhere close to Core performance (they've said that Core needs to be 2x Atom in the same power envelope, IIRC), and they need the die sizes to be very small/cheap.

I don't get what you mean that they can't "afford" Atom to be close to Core. Obviously Intel knows the tricks that got the Core IPC that fast, that R&D was already paid for. Why can't they just port some of that magic to Atom? Is the silicon that goes into Core CPUs more expensive or something?

At the same time, Atom is not the top priority for Intel; it's not going to get the resources that Core is going to get.

For example, Intel has two big Core teams -- one in Oregon, one in Haifa, this is how they do the "tick-tock." For Atom, they have just one team in Austin, TX.

This limits the scope of what the company can ultimately achieve with Atom, impacting its competitiveness in the marketplace.

The Atom team had/has talented people within its ranks, but they are not given a real chance to succeed.

I still don't understand why this Austin team can't just rip off Core designs and put the best stuff in Atom. Why can't they make Atom the movie theater in town that shows movies that are a couple of weeks old? Why are they re-inventing the wheel?

I mean I get why Apple doesn't have Core IPC magic in it, that is a bunch of Intel trade secrets. But the Atom team is the same company, why are they even having to do so much heavy lifting? You figure Atom would be at least around Ivy levels by now even without Tick Tock if they could just rip off old Core designs. I mean the guy who left for Apple has to come up with everything from scratch right?

 

[Shaun_Brannen]

I still don't understand why this Austin team can't just rip off Core designs and put the best stuff in Atom. Why can't they make Atom the movie theater in town that shows movies that are a couple of weeks old? Why are they re-inventing the wheel?

What do you mean? They can't just take everything from Conroe and bolt it onto Atom... they have power and area budgets to worry about.

 

[Shaun_Brannen]

I mean I get why Apple doesn't have Core IPC magic in it, that is a bunch of Intel trade secrets

Last I checked, they weren't far off at all.

Grr, not sure how the double post happened... whoops

 

[raghu78]

I don't get what you mean that they can't "afford" Atom to be close to Core. Obviously Intel knows the tricks that got the Core IPC that fast, that R&D was already paid for. Why can't they just port some of that magic to Atom? Is the silicon that goes into Core CPUs more expensive or something?

I still don't understand why this Austin team can't just rip off Core designs and put the best stuff in Atom. Why can't they make Atom the movie theater in town that shows movies that are a couple of weeks old? Why are they re-inventing the wheel?

I mean I get why Apple doesn't have Core IPC magic in it, that is a bunch of Intel trade secrets. But the Atom team is the same company, why are they even having to do so much heavy lifting? You figure Atom would be at least around Ivy levels by now even without Tick Tock if they could just rip off old Core designs. I mean the guy who left for Apple has to come up with everything from scratch right?

The reason Intel cannot let Atom get too close to Core is ASP and gross margins. Atom sells for cheap as it has to compete with ARM chips. Getting Atom too close to Core would cannibalize the higher margin Core sales in tablets, notebooks and even desktops. Atom has to be vastly slower for Intel's Core to be able to command such a huge price difference.

As for Apple their custom ARMv8 core in A9/A9X called Twister is very close to Intel's big cores in performance. Apple will close that gap wrt Intel in a couple of generations. I would say the A12/A12X which will be manufactured at TSMC 7nm in 2018 will be the chip which will really go after Intel's big cores like Cannonlake.

 

[poofyhairguy]

What do you mean? They can't just take everything from Conroe and bolt it onto Atom... they have power and area budgets to worry about.

Sure but they don't have to keep an artificial gap between the two (the 2X thing). Move over what magic you can move over.

Last I checked, they weren't far off at all.

Which is why Atom performance is so disappointing.

 

[poofyhairguy]

The reason Intel cannot let Atom get too close to Core is ASP and gross margins. Atom sells for cheap as it has to compete with ARM chips. Getting Atom too close to Core would cannibalize the higher margin Core sales in tablets, notebooks and even desktops. Atom has to be vastly slower for Intel's Core to be able to command such a huge price difference.

I have heard that before, but I can't mentally reconcile that with Intel spending so much marketing money to get mobile design wins. Either Intel thinks it needs to compete in mobile or it doesn't.

I mean, Intel is full of smart people. They have to know if they don't put out the best products they can eventually ARM will eat their lunch everywhere like it has in mobile.

 

[NTMBK]

I have heard that before, but I can't mentally reconcile that with Intel spending so much marketing money to get mobile design wins. Either Intel thinks it needs to compete in mobile or it doesn't.

I mean, Intel is full of smart people. They have to know if they don't put out the best products they can eventually ARM will eat their lunch everywhere like it has in mobile.

Smart people can still be dumb collectively Any company that size isn't one unified entity. It's more like a bunch of feuding fiefdoms. Mobile spend $$$ to buy market share, but the Big Core teams don't want them treading on their toes and killing their cash cow. Different teams with different agendas, and only one can win.

 

[raghu78]

I have heard that before, but I can't mentally reconcile that with Intel spending so much marketing money to get mobile design wins. Either Intel thinks it needs to compete in mobile or it doesn't.

Intel's emphasis in mobile has fallen in recent times. There are others who feel the way you do that Intel should shape up or ship out of mobile

http://www.fool.com/investing/general/2016/02/11/intel-corp-should-wind-down-its-mobile-efforts.aspx

I mean, Intel is full of smart people. They have to know if they don't put out the best products they can eventually ARM will eat their lunch everywhere like it has in mobile.

Eventually ARM will encroach all computing markets from phones/tablets to notebooks/desktops to workstations/servers. It will happen over the next decade and Intel's monopoly will be broken.

 

[IntelUser2000]

Eventually ARM will encroach all computing markets from phones/tablets to notebooks/desktops to workstations/servers. It will happen over the next decade and Intel's monopoly will be broken.

This black and white view will probably exist 25 years from now.

 

[Exophase]

Sure but they don't have to keep an artificial gap between the two (the 2X thing). Move over what magic you can move over.

A good CPU design is very balanced and synergistic. Much of the individual components simply don't make sense without the design balance of the other individual components. For example, as you increase execution width you need more reordering resources, prefetching, memory disambiguation, etc. If you have too much of those things your net efficiency goes down instead.

It all has to come together in unison. And the overarching ideal for that unison will be dictated by power and area targets.

I don't think there's really an artificial gap. It's easy to compare Atom to Twister and say that it sucks. But they have different design targets because Apple is only putting the CPU in expensive, premium flagship products. So they can afford to have something bigger, they can afford to limit themselves to two cores with big caches. These are things that those targeting the broad Android and Windows markets can't do, especially when, like Intel, they're also putting the chip in low end 28nm TSMC dies.

And it's not at all a given that Intel can do everything Apple does exactly as good or better. They have their own legacy of experience and their own bag of tricks. Intel are industry leaders in CPU design but their long term expertise has been in meeting laptop, desktop, and server targets while their lower power stuff has largely been taken for granted - in the UMPC and netbook days OEMs took what Intel would give them and like it because they were the only choice (or VIA, but they were never going to be real competitors) Only fairly recently has Intel really started building experience in aggressive low power CPU optimization and SoC integration. It was not something they could just become the best at overnight, even given their resources.

Silvermont was decent for when it came out and the 22nm node it was on. They could have and really should have been more aggressive in releasing a much better successor for 14nm. Unfortunately they're stuck in a tick-tock mindset with CPUs, where on the desktop side this looked extremely aggressive, but on the mobile side has been falling behind Apple and ARM. They need more continuous improvement to compete.

But I think they've lost motivation because even when they have a reasonably competitive SoC it's been hard to get really big wins with Android, especially phones. A lack of the connections other SoC makers have and being tied to x86 has held them back and now they might just feel like it's not worth trying anymore.

 

[know of fence]

You've posted some nice power draw graphs, but they don't do anything to touch on performance.
You can't discuss performance per watt with only half the data.
edit: Going to leave this, but after I typed it, I saw the kJ graph.

Looking at the OP, it seems that if the data were normalized for clock speed, the results might look a bit different.

Normalizing becomes non trivial when dependencies are unknown or non linear. I'm going to take another crack at interpreting numbers. Because this is by far the best benchmark, the only one that incorporates efficiency. You know, the thing that actually drives progress.

Keeping things running at 4 GHz, assuming all CPUs run base speed during heavily multi threaded encode offers a great perspective, the 32nm FX-8370 also runs at 4 GHz and needs exactly twice the energy to complete encoding.

FX-8370(32nm) to 4790K(22nm) consumption is halved!

We might see a double digit kJ value for the 32nm 2600K but it bottoms out at a meager though power efficient 3.4 GHz. We can't just arithmetically bump Sandy 2600K to 4 GHz, because the power consumption curve is bent. Normalized (incorrectly) linearly it would be 10.35 kJ @ 4 GHz, realistically it may be even higher [edit] but probably not as high as the 8 integer cores of the FX crunching x264.

Going from 32nm to 22nm gives a huge efficiency improvement, power consumption is basically halved. It's like doubling your mileage with your car.
22 nm to 14 nm decreases power by 3% (or 0.2 kJ) which is like increasing your car's mileage by 3%. We see improvement in the first place because Skylake excels at x264 encoding (9% increased IPC over HW) finishing faster with the task but running at higher wattage. In other tasks that saw no or little IPC improvement it's less power efficient than Haswell even.

We still need to determine by plotting the power curves of HW and SL when it all goes sideways, when do the steams cross...
It would also be interesting to quantify the efficiency improvement at 1.0 and 2.0 GHz, it won't be 50% reduced power, that's for sure.

 

[Abwx]

nm

 

[CHADBOGA]

Actually, Bay Trail had an integrated ISP and all of the things (sans modem) that you'd need in a modern SoC aside from connectivity.

Bay Trail's problem was that the motherboard designs that Intel provided around the SoC were very expensive. They had to do a lot of work with Bay Trail-CR to reduce the platform/board costs for low-cost tablets, and even that work wasn't enough to eliminate contra-revenue (but it took it from something along the lines of $20/unit to maybe $10/unit avg).

Cherry Trail probably brought that down further, but IIRC it was Broxton & SoFIA that were supposed to end contra-revenue in their respective market segments.

Even if we accept that Intel is intentionally depriving Atom of resources(as opposed to not having employees with a clue about mobile) to protect their cash cow in Core, the fact that they produced such inappropriately priced motherboards for Atom, shows a level of dysfunction within Intel's Mobile Group.

If they were worried about Atom powered tablets replacing laptop sales, why have Atom at all?

By putting out a hobbled Atom, they lost billions in mobile and probably made a lot of software developers become convinced that in 10 yrs or less, ARM would be able to move up the foodchain, due to how poorly x86 did against it in mobile, so some of these developers would be making long term plans to port their software to ARM.

 

[crashtech]

@know of fence, all I was really wondering is if Broadwell would stay on top if all the clockspeeds were the same.

 

[coercitiv]

Vitrol aside, we still need to determine by plotting the power curves of HW and SL when it all goes sideways, when do the beams cross...
It would also be interesting to quantify the efficiency improvement at 1.0 and 2.0 GHz, it won't be 50% reduced power, that's for sure.

I can contribute with data for i7 4510U, i7 4700HQ, i5 6600k. PM me if you need anything from these 3, I'll try to gather data through the weekend.