THIS is what Intel needs to do to get into Phones...

[extide]

So, been thinking about this for a little while. We all know that Intel has been trying to get Atom into phones for a long time. When the silvermont cores first came out they were pretty competitive, though as time has worn on they have dwindled back to being simply mediocre. They need something different, fast, and with a small enough die size to be competitive.

Let's take a look at Core m for example. Now, no, I am not saying they should try to put Core m as it is into phones, there are quite a few reasons why that both wouldn't work, and would go against Intel's current pricing and market segmentation strategy. We all know Intel wants Core m to be THE PREMIUM Ultra-Mobile x86 tablet/pc processor. It arguably is, but well it's way too expensive to go into a phone, I just can't see intel chopping the multi hundred dollar price down to phone SoC level without changing anything else. It also has several things phones don't need (bunch of PCIe, lots of USB ports, etc), and is missing a lot of things they DO need. Also, Core m is not truly a SoC. It requires the companion PCH chip that is on-package. That really won't fly in a phone these days.

So, what I believe they need to do, and frankly I would be surprised if they are not doing this already, is this. Take the Atom SoC, you know with stuff like SDIO, SPI, eMMC, all those other little interfaces you tend to see on ARM SoC's that are heavily used in phones and tablets, but rip out the Atom cores. Take a pair of Skylake cores, take out the ring bus, etc. Take just those 2 cores, and then do a large shared LLC, and pop them in, now bring along enough Gen 9 Intel GPU EU's to be competitive (in both performance and die size) as well. We end up with essentially the un-core of an Atom SoC, which is much more suited to phones than a Core m, along with 2 of the best available x86 cores, a decent enough GPU, and now they need to make sure they either put a modem on die, or if they can't manage that they need to bundle one in for cheap. Not sure how the 14nm process is with that sort of analog stuff, yet.

BOOM we end up with great compute, plenty good graphics, phone/tablet-style connectivity, you know very little SATA/PCIe, only maybe 2-3 USB 3.0 ports, those little interfaces I mentioned above, etc. As it is now the Core m (skylake) die is like 98mm^2 -- I just did some quick poking around, but I can't find the Merrfield or Cherry Trail die sizes. If anyone knows what they are I would love to know. I would think they are in the 60mm^2 range?? But anyways, I think that this new chip would still end up being similar in size to a Core m at <100mm^2 which is definitely in the range of other smartphone chips. They would need to price it at the higher end of the range for mobile SoC's ($40-50, I think, maybe a tad more) but the performance should be there to do it.

Now, that as it is would be a great chip, but what if we wanted something EVEN COOLER

Well, lets see, Intel could take it's older in-order Atom core, modernize it a bit, and put a few of those on the die too. big.LITTLE Intel Style... (Although maybe Intel can get just as much efficiency with Duty Cycling the 'big' SKYL cores)

Come on Intel MAKE IT We all know that Atom isnt going to win it for ya Intel, you need to step your game UP.

 

[poofyhairguy]

No offense, but people have been saying this for years. "Give us Core IPC at Atom prices!!!"

And Intel hasn't done it because it would cannibalize profitable segments of their business. Core IPC at Atom prices (even if it is only two cores) is basically all most of the world needs for computing power. It would eat their midrange processors alive, which means tons of margin goes down the drain. It would have been selling out the very profitable now for a speculative mobile future.

Now even that move is too late IMHO, Intel held their punches for too long and I don't think anything will get them back into mobile as a premium SoC vendor. At least not at the margins they want, the whole point of mobile is a race to the bottom to get the next billion internet users online. Intel is already in phones and no one really cares because smartphones are a saturated market and the emphasis now is the "Internet of Things" (aka SoCs and radios in everything for even LOWER margins).

If Intel is going to continue to have traction in the consumer space, then VR needs to blow up. Otherwise they might become the next Sparc, aka just a server CPU.

 

[extide]

So, I think one of the keys to this chip, would have to be that is is pretty neutered down in way that would make it not work well in PC's. Stripping out nearly all of the PCIe lanes, getting rid of SATA, etc. I kind of hinted to that above but forgot to really get into it, but yeah, that is definitely a concern. However, I think it can be mitigated. If they can remove the right stuff so it's still a great phone/tablet SoC but just not really useable in PC's then it won't cut into the rest of their sales of PC chips.

As far as budget SoC's they have that meditek partnership, they can keep on pumping out chips for cheap over there.

EDIT: Yeah I know the gist of this has been mentioned before but I was just thinking about it last night -- and wanted to put it out there.

EDIT2: It's funny though, I am still of the opinion that they never should have ditched the Intel XScale chips, not sold them to Marvell, and never made the Atom at all, and just kept making custom ARM IP. I mean yeah they are famous for IA chips but that doesnt mean they have to be the only thing they do. I think just the fact that it would be an ARM chip as opposed to not ARM it would have ultimately been very successful -- you know more palatable to OEMS. I mean, for those who remember the old windows mobile days, those Intel XScale chips were in pretty much EVERY phone. They could have just kept revving that and been in a totally different place than they are today. As good as Intel is, they sure have made quite a few great mis steps (P4, Itanium, XScale, ...)

 

[VirtualLarry]

Correct me if I'm wrong, but isn't "Core M" really just Core, tuned for slightly lower power consumption? I mean, if Core M won't scale down to the power usage levels that a phone SoC demands, what makes you think Core will scale lower?

 

[aigomorla]

*looks at the battery size for an atom netbook"

Compares to Size of battery on my S6 Edge....

:\

uhh... no i think i'll stick to SoC for my cell phone.

 

[poofyhairguy]

So, I think one of the keys to this chip, would have to be that is is pretty neutered down in way that would make it not work well in PC's. Stripping out nearly all of the PCIe lanes, getting rid of SATA, etc. I kind of hinted to that above but forgot to really get into it, but yeah, that is definitely a concern. However, I think it can be mitigated. If they can remove the right stuff so it's still a great phone/tablet SoC but just not really useable in PC's then it won't cut into the rest of their sales of PC chips.

But that is the problem- the market determines what happens with Intel's products more than Intel. If they put out some mobile competitor that is basically just as powerful as their $100+ desktop chip for $15 (which is right where mobile is) then someone like a Lenovo is going to take that chip and stick it in a motherboard with an old-school southbridge that brings back the ability to add sata drives or whatever is needed to make a crappy cheap laptop work. If Intel cuts out too much from the chip to try an make it unappealing for a laptop, eventually it will be unappealing for mobile too.

Intel didn't lose the war on mobile because they don't have good chips or because they can't compete. They lost mobile because they REFUSED to move towards a business model with lower margins and so their SoCs have lacked either IPC or GPU power from day one.

EDIT2: It's funny though, I am still of the opinion that they never should have ditched the Intel XScale chips, not sold them to Marvell, and never made the Atom at all, and just kept making custom ARM IP. I mean yeah they are famous for IA chips but that doesnt mean they have to be the only thing they do. I think just the fact that it would be an ARM chip as opposed to not ARM it would have ultimately been very successful -- you know more palatable to OEMS. I mean, for those who remember the old windows mobile days, those Intel XScale chips were in pretty much EVERY phone. They could have just kept revving that and been in a totally different place than they are today. As good as Intel is, they sure have made quite a few great mis steps (P4, Itanium, XScale, ...)

I think you miss the point of the ARM revolution. It didn't happen because ARM SoC makers have really good chips (some are good, some are crap), or because the ARM design is so superior to anything else.

ARM won BECAUSE of the ease of getting a licence. With x86 three companies could make chips, so only three companies could compete in that market. With ARM anyone who can afford a cheap licence can start making their own CPUs and because of that we have seen a rush of new SoC makers, each willing to take a little less margin than the last SoC maker.

Competition in the mobile market has made that market what it is. Intel's current business model couldn't survive in a market with those kind of margins, and quite frankly even if they would have had the vision to try and compete way back when the stockholders wouldn't have let them throw away high margins for volume. The whole contra thing was them realizing that their stubbornness about margin was making them lose the mobile market before they ever really got into it, so they tried the tactics that worked oh so well in the Wintel era (paying off OEMs, pressure on distributors, even paying for a chunk of x86 Android development). But in a more competitive market those tactics didn't work, and they don't have one major smartphone design win to date (major as in sold in a carrier store). Even worse, their best customer in the consumer space (Apple) is now making their own very competitive CPUs.

The walls are closing in for Intel and mobile, that is why they are focusing on robots.

 

[Th3Loonatic]

So, been thinking about this for a little while. We all know that Intel has been trying to get Atom into phones for a long time. When the silvermont cores first came out they were pretty competitive, though as time has worn on they have dwindled back to being simply mediocre. They need something different, fast, and with a small enough die size to be competitive.

Let's take a look at Core m for example. Now, no, I am not saying they should try to put Core m as it is into phones, there are quite a few reasons why that both wouldn't work, and would go against Intel's current pricing and market segmentation strategy. We all know Intel wants Core m to be THE PREMIUM Ultra-Mobile x86 tablet/pc processor. It arguably is, but well it's way too expensive to go into a phone, I just can't see intel chopping the multi hundred dollar price down to phone SoC level without changing anything else. It also has several things phones don't need (bunch of PCIe, lots of USB ports, etc), and is missing a lot of things they DO need. Also, Core m is not truly a SoC. It requires the companion PCH chip that is on-package. That really won't fly in a phone these days.

So, what I believe they need to do, and frankly I would be surprised if they are not doing this already, is this. Take the Atom SoC, you know with stuff like SDIO, SPI, eMMC, all those other little interfaces you tend to see on ARM SoC's that are heavily used in phones and tablets, but rip out the Atom cores. Take a pair of Skylake cores, take out the ring bus, etc. Take just those 2 cores, and then do a large shared LLC, and pop them in, now bring along enough Gen 9 Intel GPU EU's to be competitive (in both performance and die size) as well. We end up with essentially the un-core of an Atom SoC, which is much more suited to phones than a Core m, along with 2 of the best available x86 cores, a decent enough GPU, and now they need to make sure they either put a modem on die, or if they can't manage that they need to bundle one in for cheap. Not sure how the 14nm process is with that sort of analog stuff, yet.

BOOM we end up with great compute, plenty good graphics, phone/tablet-style connectivity, you know very little SATA/PCIe, only maybe 2-3 USB 3.0 ports, those little interfaces I mentioned above, etc. As it is now the Core m (skylake) die is like 98mm^2 -- I just did some quick poking around, but I can't find the Merrfield or Cherry Trail die sizes. If anyone knows what they are I would love to know. I would think they are in the 60mm^2 range?? But anyways, I think that this new chip would still end up being similar in size to a Core m at <100mm^2 which is definitely in the range of other smartphone chips. They would need to price it at the higher end of the range for mobile SoC's ($40-50, I think, maybe a tad more) but the performance should be there to do it.

Now, that as it is would be a great chip, but what if we wanted something EVEN COOLER

Well, lets see, Intel could take it's older in-order Atom core, modernize it a bit, and put a few of those on the die too. big.LITTLE Intel Style... (Although maybe Intel can get just as much efficiency with Duty Cycling the 'big' SKYL cores)

Come on Intel MAKE IT We all know that Atom isnt going to win it for ya Intel, you need to step your game UP.

It's non trivial to just remove the Atom cores from the SOC and put in a big Core. It's likely that those IOs have been designed with Atom in mind. Putting a big Core in it would require a whole new design and at that stage you might as well be making an entire new product with all the design and validation work it entails.

 

[StrangerGuy]

OP is thinking too much like a traditional PC enthusiast, when the truth is once mobile performance and features reach a good enough level, it all boils down to cost, cost and cost. Which is why ARM has won the mobile race before Intel crossed the starting line.

Even with the ARM's explosion of performance, almost nobody has even bothered to utilize it. The most popular games on Android are still 2D, much less remotely tax a Adreno 320 released back in 2011, while actual taxing compute stuff like video encoding/decoding are already hardware-accelerated by SoC DSPs.

 

[Vesku]

No offense, but people have been saying this for years. "Give us Core IPC at Atom prices!!!"

And Intel hasn't done it because it would cannibalize profitable segments of their business. Core IPC at Atom prices (even if it is only two cores) is basically all most of the world needs for computing power. It would eat their midrange processors alive, which means tons of margin goes down the drain. It would have been selling out the very profitable now for a speculative mobile future.

Now even that move is too late IMHO, Intel held their punches for too long and I don't think anything will get them back into mobile as a premium SoC vendor. At least not at the margins they want, the whole point of mobile is a race to the bottom to get the next billion internet users online. Intel is already in phones and no one really cares because smartphones are a saturated market and the emphasis now is the "Internet of Things" (aka SoCs and radios in everything for even LOWER margins).

If Intel is going to continue to have traction in the consumer space, then VR needs to blow up. Otherwise they might become the next Sparc, aka just a server CPU.

Otellini basically said this when he was leaving, Intel management didn't want to take the gross margin hit necessary to get into the first iPhone. Imo, same reason Intel went with a rebate scheme with Bay Trail. Hides a bit from shareholders the reduction in margins that is necessary to compete in mobile since Intel didn't secure premium branding positions at the start of the smartphone era.

Imo, it's why the current CEO is pursuing a bit of a "try everything" approach. ARM will eventually be a real threat to Intel's traditional low end PC market. So they really want to find a promising new market where they can once again leverage their foundry resources to dominate the premium space.

 

[CHADBOGA]

Yes Intel should find ways to break the laws of physics.

 

[turtile]

You're forgetting the most important part of the chip, the communication components. It's hard to compete with Qualcomm when they can give you every modem etc. altogether (and very good to top it off).

 

[Exophase]

Otellini basically said this when he was leaving, Intel management didn't want to take the gross margin hit necessary to get into the first iPhone. Imo, same reason Intel went with a rebate scheme with Bay Trail. Hides a bit from shareholders the reduction in margins that is necessary to compete in mobile since Intel didn't secure premium branding positions at the start of the smartphone era.

Imo, it's why the current CEO is pursuing a bit of a "try everything" approach. ARM will eventually be a real threat to Intel's traditional low end PC market. So they really want to find a promising new market where they can once again leverage their foundry resources to dominate the premium space.

Intel was nowhere close to ready to compete with Samsung, TI, Qualcomm and other SoC manufacturers when the smart phone revolution was kicked off by iPhone. While the first Atom CPUs had a reasonable design for this purpose it took years for the peripheral integration, low power process modifications, and power management to catch up.

 

[Vesku]

Intel was nowhere close to ready to compete with Samsung, TI, Qualcomm and other SoC manufacturers when the smart phone revolution was kicked off by iPhone. While the first Atom CPUs had a reasonable design for this purpose it took years for the peripheral integration, low power process modifications, and power management to catch up.

I imagine licensing from ARM or buying out an existing ARM SoC company was not out of the question when Otellini was having tentative discussions with Jobs. They would have definitely had to buy a 3G modem company.

 

[ninaholic37]

Intel is trying to do the opposite. Atom in 15.6" laptops, Atom in desktops. I think they said they wanted to use Atom for their next supercomputer.

 

[JonFX01]

Stop giving them ideas D:

 

[NTMBK]

Even Cortex A57/A72 isn't that popular in phones. The high volume mid-market Android phones all use A53 cores, and larger SoCs do most of their work on the smaller A53 cores, only firing up the big cores in rare cases. I really don't see the market for a Skylake phone. Maybe on the 10nm shrink?

 

[Thala]

There is potentially nothing wrong with only being a high-end SoC provider. However Atom is not remotely competitive in this space. And with Atom pushed into mid-range or even low-end with SoFIA the profit margins are gone.

Intel is trying to do the opposite. Atom in 15.6" laptops, Atom in desktops.

Its more the OEMs trying to do so given the Atom is currently available at a big discount (aka counter revenue). It is much less in the interest of Intel.

I also would agree on the XScale statement. XScale was the leading ARM SoC back in the days and who knows how XScale would look today...given Intels technology leadership.

 

[lopri]

Often we ridicule Intel's mobile SOCs in comparison to other mobile SOCs riding on different CPU architecture, but IMO the GPU portion of Atom has been a bigger problem.

 

[podspi]

Often we ridicule Intel's mobile SOCs in comparison to other mobile SOCs riding on different CPU architecture, but IMO the GPU portion of Atom has been a bigger problem.

Yes. There are mass market phones with Intel SoCs in them. The GPU performance is always terribad, it's like they can't help themselves.

 

[frozentundra123456]

There is potentially nothing wrong with only being a high-end SoC provider. However Atom is not remotely competitive in this space. And with Atom pushed into mid-range or even low-end with SoFIA the profit margins are gone.



Its more the OEMs trying to do so given the Atom is currently available at a big discount (aka counter revenue). It is much less in the interest of Intel.

I also would agree on the XScale statement. XScale was the leading ARM SoC back in the days and who knows how XScale would look today...given Intels technology leadership.

It's "contra" revenue, and there is none for laptops and desktops. Atom is just a cheap chip compared to core.

 

[sm625]

You wouldnt even need multiple skylake cores. Even just one would suffice. One skylake and 2 atom and hell throw in a couple quarks just to round it out. big.LITTLE.tINy. One hyperthreaded skylake core would crush mobile apps.

 

[Nothingness]

You wouldnt even need multiple skylake cores. Even just one would suffice. One skylake and 2 atom and hell throw in a couple quarks just to round it out. big.LITTLE.tINy. One hyperthreaded skylake core would crush mobile apps.

It would also crush your battery. And if you start limiting frequency to say 1GHz to reduce power consumption (and even that frequency is perhaps not low enough to make it fit in a smartphone), suddenly the CPU won't crush mobile apps any more. Intel has no more magic dust than ARM, there's a reason Intel designed a specific CPU

 

[Exophase]

Even Cortex A57/A72 isn't that popular in phones. The high volume mid-market Android phones all use A53 cores, and larger SoCs do most of their work on the smaller A53 cores, only firing up the big cores in rare cases. I really don't see the market for a Skylake phone. Maybe on the 10nm shrink?

It's way too early to talk about A72's popularity or lack thereof. Given the big appearance it'll have in market slots where A57 wasn't for Qualcomm and MediaTek I wouldn't write it off its mainstream appeal.

I also think "rare cases" is an overstatement for current big.LITTLE phones, but it depends a lot on what you run (for example, if you're trying to run something like Dolphin you're going to really want those big cores working as hard as they can)

 

[extide]

So most of you guys seem to be of the opinion that the battle is lost and they will never get into phones eh? Not even try?

What chip would you build if you were Intel right now, and trying to get into phones?

 

[aigomorla]

So most of you guys seem to be of the opinion that the battle is lost and they will never get into phones eh? Not even try?

What chip would you build if you were Intel right now, and trying to get into phones?

NAND and RAM related stuff....
since everything uses one of them...
But they already do that under a partner called micron.

And no i dont believe intel has given up.
They are more "playing" with products / competitors.

I consider this stage simular to what happened against AMD back in the Athlon days.
P4 was a horrendous product, then intel threw out the C2D and said GG.

Intel is waiting for another C2D from their sandbox.
I expect a ticking time bomb before intel goes GG again, especially since they are the only chip company which can dump 1 billion dollars on a heart beat to draft another fab, if deem'd necessary, on the latest node even.