Doesn't it make sense to fab these simpler mobile chips on the newest process node - to iron out issues and to get a reasonable yield sooner - instead of a more complex Core ix chip?
We always make SRAM cells on a new process node at first. They're easy to make and highly repetitious in design, so it makes finding process flaws easier.
Doubt it's about the cores or the process, although the process might be a factor especially after all the 14nm problems. More likely it's about all the other compute units on the chip since those can be used in everything and included or removed based on market they address. After all in something like the Snapdragon 800, the cores are less than 20% of the die, cores plus GPU maybe 1/3 of it ( i think it was die size 118mm2, cores 22mm2, GPU 16mm2, LTE 26mm2).
Regardless of the design, going to 14 nm and beyond is going to be very painful for all foundries. 10 nm seems solid on Intel's roadmap and they can likely do 7 nm, I'm very pessimistic beyond that point.
Hm. You're very optimistic. Intel only has so many resources even if they are as rich as sin. Something has to give. If they put mobile on equal footing with their "real" chips, then their real chips will suffer some loss in attention as Intel pivots to try and focus more fab time on their mobile chips.
Intel is already delaying chips at every release. That's probably the logic here, of course. AMD is no competition at all, so they can delay their releases with impunity and lose nothing. The only competition they have in the mid-range and high end are essentially their own releases from yesteryear.
So shifting resources CAN work as long as they don't let chasing mobility tear too much attention away from the real meat and potatoes of their day to day. Microsoft's proven that it's foolish to chase mobility at the expense of your main line, so Intel should not make the same mistake now.
In truth, Intel could ignore mobility entirely and continue their progress with Core, pushing it lower and lower. That'd wind up where pushing Atom up will and it'd probably be far cheaper in the long run. And that might be yet another reason to merge the divisions. This might already have happened for the products to come in 2016 or so.
Regardless, I'm hoping Intel ignores the mobility fad focus that other companies are doing (and failing at). Let AMD and Microsoft make that mistake. Focus on your Core business and just bring it down to mobile levels slowly and surely just like you already are.
Then by the time mobility gets up to reasonable performance levels required to do more than view content (ie., web browse, read emails, watch Netflix, watch youtube), Intel will be ready with chips that are both powerful and reasonably low power.
Try to rush to the end with chips that sacrifice performance and you'll get huge losses.
Just like Intel's getting right now with their mobility division the way it is. Probably best to just dump the entire Atom line and be patient until Core can be fit into lower power specs. Sometimes a problem can't be fixed right away and patience is required.
Or Intel can throw money away on trying to peddle weak-by-comparison Atom into devices no one wants it in.
My initial thought on this is the reason why Intel would want/need to rush into the mobile market is to keep x86 relevant there. The more time ARM has to own the market, the more entrenched that architecture will become. Similar to how if ARM had a chip as fast as an i7 (which they obviously don't) it wouldn't even be a consideration for a desktop or laptop as it can't run any of the software people want on a desktop or laptop (Windows, Office, etc). The longer Intel lets ARM run the tablet space the more apps will be written for ARM and the harder it'll be for them to get into it.
Maybe. But at least with Android many apps are CPU agnostic since they run via a runtime, and we already have ARM and Intel devices on the market. Even Windows has this capability with .NET... some .NET apps can run on both x86 and ARM with no modification. I can't speak for iOS though...
Most Android apps, or at least games, are partially in ARM C. For this reason, Intel had to write an ARM interpretor (libhoudini) for x86 Android devices.
Apple cares about power more than anything else because they see power as enabling everything. Intel claims to believe this, but they still don't get it. Look at their crappy Quark chip.
Right now we're at the equivalent of 2007 when it comes to wearables and Internet of Things --- we know they're about to explode, we're all waiting to see what Apple has cooked up in the labs. And Intel, rather than offering a seriously competitive chip has offered us some PoS 486 derivative. Seriously? How many watch wins has that thing got? How many they expect it to get.
By the time the dust has settled and IoT is real (so maybe 5 years from now) we'll all be familiar with and arguing about the merits of the various ARM ultra-low power processors, GPUs, and wireless chipsets; and Intel will be once again whining about the fact that no-one cares about their stupid Quark version 3 chip, now upgraded to all the sophistication of a Pentium and low enough power to fit in a phone (not yet a watch, but just you wait, guys the next rev will be awesome...)
So, yeah, I expect Apple have iOS running on Intel (including full XCode) and could flip the switch any time they needed to. I also expect it won't ever come to that because Intel can't find their ass with a flashlight and are constantly planning to fighting last year's war two years from now. By the time they FINALLY have a chip that anyone sane would want to use in a watch, Apple will have move on to considering just what features they should add to the iDust motes that they're working on.
I don't think that will work. When a new line of products comes you expect it to be better then before. If they stick to the 100W CPU-s they are making for the desktop they will never make a 2W CPU. In 20 years 100W will still be 100W and your phone won't be able to handle that no matter what. If they start lowering power draw and performance their products will be weaker than those from AMD and they will lose the desktop and never gain the mobile.
The issue is not fab capacity, it is design capacity. How many world-class chips can they actually design? Remember that to design a Core CPU takes at least 7 years. They generate one a year (or used to, until Broadwell...) by having 7 parallel design teams. The Atom upgrade took so long because that was one team which, after the shipped the first one, needed what, five years or so to create the next one. So are they going to tool up for six or seven parallel Atom teams --- while Atom is still losing money? Also some number of Xeon teams (I don't know how many) working mainly on uncore . Meanwhile there's at least one Phi team which is burning money on a product that has no obvious long term money making potential. And there's pathetic little Quark in the background --- at some point (too late) they'll have to introduce something competitive in that space.
Even if you're Intel, it's hard to FIND that many good people and to pay them. And face it, if you're a hot shot designer, wouldn't you rather work at Apple or nV or ARM, somewhere where 90% of your time is spent on new ideas, rather than at Intel where 90% of your time is spent on stupid crap to ensure compatibility with some crazy design decision from 1982?
"Microsoft's proven that it's foolish to chase mobility at the expense of your main line, so Intel should not make the same mistake now." Microsoft deals in software and user interface, not hardware. NVIDIA went the mobility over performance route, and look at their current line of GPUs. The 980 has the highest efficiency of any GPU, and the 980m brings incredible performance to laptops.
I think that Intel focusing on mobile together with Core processors would help get the heat down, and then maybe we'll start seeing 5+ GHz stock clocks on desktops.
You're missing the larger point. Intel is facing classic Innovator's Dilemma. Their heart is in making ever fancier, ever faster Core CPUs. BUT for most purposes now, most of that sophistication in a Core CPU, starting from the x86 ISA down, is not essential --- which means here is no compelling reason for customers to pay $300 for those features, when they can get an ARM that does the same job at $30.
Intel kinda sorta knows it has to deal with this, but they haven't figured out a way. They want to apply the same segmentation crap they have used before, but it's no longer working. There was a time, for example, when if you wanted 4 cores, you'd have to pay desktop money --- no Intel 4 cores for your low-end netbook. Problem is, plenty of ARM vendors are willing to sell a 4-core system. Likewise for virtualization. Intel can try to keep those sorts of enterprise features out of their low-end, but turns out that some of them (things like TrustZone, or crypto instructions) are so essential that ARM ALREADY does them better than Intel, and the ARM vendors targeting servers are busily solving the problems of ARM virtualization.
So Intel is kinda screwed on all dimensions. If they keep their Atoms crappy, fine, everyone will buy ARM --- Intel has nothing special in their Atoms that make them compelling unless you REALLY need x86 compatibility. But if they improve their Atoms, then those Atoms become good enough for many of the purposes that used to require a CPU at 4 or 5 times the cost.
As if that weren't bad enough, they're facing the same problem on the server side. It's not obvious yet, but in a year or two it will be. Once again ARM vendors are willing to sell the same thing as Intel at half the cost. The ONLY advantage Intel has is x86 ISA, and that's becoming less and less important by the month. Once again Intel wants to limit all its weirdly acronymed Enterprise features to the expensive Xeons, but ARM vendors will give you those same features as part of the package on their low-end server chips.
Few companies have handled this sort of transition gracefully. Apple has --- they were willing to let iPhone cannibalize iPod, and iPad to cannibalize MBA (if in fact this happens --- some claim it does, though I doubt it). But for Intel it means having to match ARM prices, which is likely impossible the way Intel is currently set up. Intel probably COULD become the dominant ARM vendor if they were willing to accept the ARM ISA. But they are committed to x86, and the cost that that imposes in design overhead and verification makes it economically impossible for them to compete. It'll be very interesting to see how their pricing structures and product segmentation evolve over the next two years...
Your arguments make sense if cost is the only concern. They fall apart if you have any desire at all for performance. Intel's price/performance ratios can't be beat in servers. Silvermont is also better perf/watt than current Qualcomm cores. The rest of the SOC (Baytrail) is not so great, but the CPU cores are very good.
I doubt that Intel wants to compete with AMD at this point. The x86 cpu market has only 2 players left and if Intel decides to squeeze AMD out of the market the Anti-trust police will come after Intel. In fact, Intel is not doing anything about AMD into the APU market and this market is going to be stagnant.
Intel has severely miscalculated the mobile market early on and now is paying the price. Intel always been selling of ditching divisions that is not making money just because of short term reasons like getting rid of their StrongARM business years back. Even so, Intel is doing things half-assed in their mobile dept because they are making chips in 22nm when they can make it at 14nm. Intel should be thinking of putting their chips on cell phones where there is a bigger profit margin compared to tablets where they can't compete of the likes of allwinner.
Actually the Core M so far seems to be far worse than the Bay Trail Moorefield, considering the price and process.
If Intel wants to compete in the ULP space, I think they need to get rid of the desktop legacy arhitecture actually. Shrinking, restraining and restricting the desktop architecture bellow 5W TDP is not the way to go and the Core M so far proves it. The fact that comparatively super cheap Apple A8X shows similar performance and far better efficiency on a less advanced process is a major blow to the myth of Intel.
I don't think A8X is showing similar performance to Core M, although I admittedly haven't seen a chart. It's an interesting point though, would Anandtech consider trying to do some kind of perf/watt benchmark? It might not be super easy, but would be really good to see.
I don't think the Atom CPU is useless... the Asus T100 is a nice little device for example. Sure it's no speed demon, but it can do a lot of useful things just fine.
Some degree of separation should be in play with the development of these processor families. It's helped Intel in the past and in some ways its reasonable to assume that it's part of the reason why they have what they do today.
This is supposed to help their mobile unit? One of Intel's problems is the lack of a radio on their mobile SOCs. How does splitting the mobile CPU and RF units help with that?
I am wondering if this is going to be used as an excuse when they bump the ship date again for the Broadwell desktop parts.
I expect this to be a better way to hide how much money their mobile division is losing. Intel needs the next big thing to drive a new era of exponential growth. The loss of Denson scaling has destroyed Intel's ablity to make its own older parts obsolete. Every 4 years a new PC was a requirement before 2004. Now a new PC is a luxury, and Mobile replaces half of the PC's market place with good enough content consumption machines.
Intel has to get photonics and a new replacement for CMOS up and runing pronto. Then they can afford to eat their own success story. The cannot afford to let someone else destroy the PC.
To further clarify, they OoO'd the Atom and double the core count and it's still not competitive in performance. The Core is scaling down in power surprisingly well. Makes sense to kill the Atom.
It is, but it still isn't competitive with Atom in one area where it really counts for mobile.
Power draw. Core M is still higher power draw than Bay Trail-T (if not by huge amounts), its performance is not massively better (it IS better) and it deffinitely cannot be made in to a phone chip, which is part of where Intel is trying to go with Atom.
Now I could possibly see Intel refocusing Atom to be a phone only chip and do derivative Core designs to fill the tablet space from bottom to top, but I don't see them managing to make Core in to a phone chip ever. Laws of scaling, one design can really only do well with a 10x power factor. For Core, that means roughly 10-13w up to 100-130w. I'd semi-argue what Intel is really targeting is 9w-90w with the big iron really being the same architecture scaled out to extra cores. Core M makes some fairly big sacrafices to fit in to 5w, and looking at it, it appears to sacrafice performance per watt to get there. It most deffinitely does not seem to be an ideal design for that low a power consumption.
What Intel kind of needs is an even better Atom architecture that scales better from 1-10w, which I think is the direction they are moving. It is just going to take them a couple of years to get there. Airmont/Cherry Trail seems to be promising about a 15-30% increase in CPU performance (baring any unannounced major Architecture tweaks) and a 2-4x increase in GPU performance over Bay Trail. That'll GREATLY close the gap with Core M as it stands today. In fact it should blur almost all distinctions as it would likely leave Core M with only a small GPU advantage and possibly even a DISADVANTAGE in highly multithreaded applications, depending on throttling, workloads, etc.
Likely the next release of Atom, Goldmont/Willow Trail, is going to focus pretty heavily on new architecture for the CPU and might well yield some very large CPU gains.
That said, IMHO, Bay Trail makes an excellent tablet chip (I have a T100), a good netbook chip and a poor laptop chip (at least Bay Trail-T, but frankly Bay Trail-N/D isn't really that much better performing). Core M seems to make an okay laptop chip, an okay netbook chip and an okay tablet chip. In a couple more years, if Intel can keep iterating and delivering, Atom is likely to continue to make an excellent tablet chip and netbook chip and might even make an okay laptop chip.
Is Airmont still supposed to be released this year?
I think everyone is over-reacting slightly to this management news. This kind of engineering takes time. Sharing management might impact designs on either the mobile or core front, but even if it does it won't show up in selling product for a couple years at least. This is probably more an acknowledgement that at a certain point the hardware plateaus and there will be zero difference between the mobile parts and the low end desktop parts. So why have two divisions trying to design/build them? You'll just have one class of chip with the power efficiency to run mobile devices and the power capacity to service anything short of a work station. We're clearly on the doorstep of that era now.
The process Intel uses for it's server, desktop and conventional laptop processors is different than the one they use for atom laptop chips and their attempted push into mobile. The latter's tuned for lower TDP instead of higher clock rates. The way Intel's pushing on ultrabooks though; assuming they can get the low power process caught up timeline wise, I wouldn't be surprised if they move ULP Core processors over to the latter too.
It may be strange to say, but it seam that Intel is blocked by GPU designs more then CPUs. Like this new Nokia tablet with Intel CPU and... GPU from somebody else.
Either Intel price GPUs without those big incentives CPUs receive or Intel is not ready there.
From Linux work for next gen hw (hw AFTER broadwell), we can see that Intel is pursuing tilling GPUs, so they are aware of market requirements at least. Hope they are quick about it. They have some very good strong points beside performance (fully opensource cpu/gpu drivers!!!)
Baytrail is the only Intel Atom ever to have in Intel GPU. All other Atoms have had 3rd party GPU's in them, PowerVR to be precise, the same GPU Apple uses in their Ax chips. The Nokia tablet uses Moorefield which has the PowerVR G6430 GPU in it.
You're forgetting one thing. Intel's Core chips are 5-10x more expensive than Atoms. And Intel is already heavily subsidizing Atom. Core will be hilariously noncompetitive in the mobile market.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
34 Comments
Back to Article
Phylyp - Tuesday, November 18, 2014 - link
Doesn't it make sense to fab these simpler mobile chips on the newest process node - to iron out issues and to get a reasonable yield sooner - instead of a more complex Core ix chip?przemo_li - Wednesday, November 19, 2014 - link
Not if You can't get volumes.JKflipflop98 - Saturday, November 22, 2014 - link
We always make SRAM cells on a new process node at first. They're easy to make and highly repetitious in design, so it makes finding process flaws easier.jjj - Tuesday, November 18, 2014 - link
Doubt it's about the cores or the process, although the process might be a factor especially after all the 14nm problems.More likely it's about all the other compute units on the chip since those can be used in everything and included or removed based on market they address. After all in something like the Snapdragon 800, the cores are less than 20% of the die, cores plus GPU maybe 1/3 of it ( i think it was die size 118mm2, cores 22mm2, GPU 16mm2, LTE 26mm2).
Kevin G - Tuesday, November 18, 2014 - link
Regardless of the design, going to 14 nm and beyond is going to be very painful for all foundries. 10 nm seems solid on Intel's roadmap and they can likely do 7 nm, I'm very pessimistic beyond that point.HisDivineOrder - Tuesday, November 18, 2014 - link
Hm. You're very optimistic. Intel only has so many resources even if they are as rich as sin. Something has to give. If they put mobile on equal footing with their "real" chips, then their real chips will suffer some loss in attention as Intel pivots to try and focus more fab time on their mobile chips.Intel is already delaying chips at every release. That's probably the logic here, of course. AMD is no competition at all, so they can delay their releases with impunity and lose nothing. The only competition they have in the mid-range and high end are essentially their own releases from yesteryear.
So shifting resources CAN work as long as they don't let chasing mobility tear too much attention away from the real meat and potatoes of their day to day. Microsoft's proven that it's foolish to chase mobility at the expense of your main line, so Intel should not make the same mistake now.
In truth, Intel could ignore mobility entirely and continue their progress with Core, pushing it lower and lower. That'd wind up where pushing Atom up will and it'd probably be far cheaper in the long run. And that might be yet another reason to merge the divisions. This might already have happened for the products to come in 2016 or so.
Regardless, I'm hoping Intel ignores the mobility fad focus that other companies are doing (and failing at). Let AMD and Microsoft make that mistake. Focus on your Core business and just bring it down to mobile levels slowly and surely just like you already are.
Then by the time mobility gets up to reasonable performance levels required to do more than view content (ie., web browse, read emails, watch Netflix, watch youtube), Intel will be ready with chips that are both powerful and reasonably low power.
Try to rush to the end with chips that sacrifice performance and you'll get huge losses.
Just like Intel's getting right now with their mobility division the way it is. Probably best to just dump the entire Atom line and be patient until Core can be fit into lower power specs. Sometimes a problem can't be fixed right away and patience is required.
Or Intel can throw money away on trying to peddle weak-by-comparison Atom into devices no one wants it in.
HunterKlynn - Tuesday, November 18, 2014 - link
My initial thought on this is the reason why Intel would want/need to rush into the mobile market is to keep x86 relevant there. The more time ARM has to own the market, the more entrenched that architecture will become. Similar to how if ARM had a chip as fast as an i7 (which they obviously don't) it wouldn't even be a consideration for a desktop or laptop as it can't run any of the software people want on a desktop or laptop (Windows, Office, etc). The longer Intel lets ARM run the tablet space the more apps will be written for ARM and the harder it'll be for them to get into it.domboy - Tuesday, November 18, 2014 - link
Maybe. But at least with Android many apps are CPU agnostic since they run via a runtime, and we already have ARM and Intel devices on the market. Even Windows has this capability with .NET... some .NET apps can run on both x86 and ARM with no modification. I can't speak for iOS though...Mikemk - Tuesday, November 18, 2014 - link
Most Android apps, or at least games, are partially in ARM C. For this reason, Intel had to write an ARM interpretor (libhoudini) for x86 Android devices.name99 - Tuesday, November 18, 2014 - link
Apple cares about power more than anything else because they see power as enabling everything.Intel claims to believe this, but they still don't get it. Look at their crappy Quark chip.
Right now we're at the equivalent of 2007 when it comes to wearables and Internet of Things --- we know they're about to explode, we're all waiting to see what Apple has cooked up in the labs. And Intel, rather than offering a seriously competitive chip has offered us some PoS 486 derivative. Seriously? How many watch wins has that thing got? How many they expect it to get.
By the time the dust has settled and IoT is real (so maybe 5 years from now) we'll all be familiar with and arguing about the merits of the various ARM ultra-low power processors, GPUs, and wireless chipsets; and Intel will be once again whining about the fact that no-one cares about their stupid Quark version 3 chip, now upgraded to all the sophistication of a Pentium and low enough power to fit in a phone (not yet a watch, but just you wait, guys the next rev will be awesome...)
So, yeah, I expect Apple have iOS running on Intel (including full XCode) and could flip the switch any time they needed to. I also expect it won't ever come to that because Intel can't find their ass with a flashlight and are constantly planning to fighting last year's war two years from now. By the time they FINALLY have a chip that anyone sane would want to use in a watch, Apple will have move on to considering just what features they should add to the iDust motes that they're working on.
SleepyFE - Tuesday, November 18, 2014 - link
I don't think that will work. When a new line of products comes you expect it to be better then before. If they stick to the 100W CPU-s they are making for the desktop they will never make a 2W CPU. In 20 years 100W will still be 100W and your phone won't be able to handle that no matter what. If they start lowering power draw and performance their products will be weaker than those from AMD and they will lose the desktop and never gain the mobile.coburn_c - Tuesday, November 18, 2014 - link
"Intel only has so many resources"Other way around. They cancelled the opening of their giant new Arizona 14nm fab because they didn't need the excess capacity.
name99 - Tuesday, November 18, 2014 - link
The issue is not fab capacity, it is design capacity. How many world-class chips can they actually design?Remember that to design a Core CPU takes at least 7 years. They generate one a year (or used to, until Broadwell...) by having 7 parallel design teams. The Atom upgrade took so long because that was one team which, after the shipped the first one, needed what, five years or so to create the next one.
So are they going to tool up for six or seven parallel Atom teams --- while Atom is still losing money? Also some number of Xeon teams (I don't know how many) working mainly on uncore .
Meanwhile there's at least one Phi team which is burning money on a product that has no obvious long term money making potential. And there's pathetic little Quark in the background --- at some point (too late) they'll have to introduce something competitive in that space.
Even if you're Intel, it's hard to FIND that many good people and to pay them. And face it, if you're a hot shot designer, wouldn't you rather work at Apple or nV or ARM, somewhere where 90% of your time is spent on new ideas, rather than at Intel where 90% of your time is spent on stupid crap to ensure compatibility with some crazy design decision from 1982?
Mikemk - Tuesday, November 18, 2014 - link
"Microsoft's proven that it's foolish to chase mobility at the expense of your main line, so Intel should not make the same mistake now."Microsoft deals in software and user interface, not hardware. NVIDIA went the mobility over performance route, and look at their current line of GPUs. The 980 has the highest efficiency of any GPU, and the 980m brings incredible performance to laptops.
I think that Intel focusing on mobile together with Core processors would help get the heat down, and then maybe we'll start seeing 5+ GHz stock clocks on desktops.
name99 - Tuesday, November 18, 2014 - link
You're missing the larger point. Intel is facing classic Innovator's Dilemma.Their heart is in making ever fancier, ever faster Core CPUs. BUT for most purposes now, most of that sophistication in a Core CPU, starting from the x86 ISA down, is not essential --- which means here is no compelling reason for customers to pay $300 for those features, when they can get an ARM that does the same job at $30.
Intel kinda sorta knows it has to deal with this, but they haven't figured out a way. They want to apply the same segmentation crap they have used before, but it's no longer working. There was a time, for example, when if you wanted 4 cores, you'd have to pay desktop money --- no Intel 4 cores for your low-end netbook. Problem is, plenty of ARM vendors are willing to sell a 4-core system. Likewise for virtualization. Intel can try to keep those sorts of enterprise features out of their low-end, but turns out that some of them (things like TrustZone, or crypto instructions) are so essential that ARM ALREADY does them better than Intel, and the ARM vendors targeting servers are busily solving the problems of ARM virtualization.
So Intel is kinda screwed on all dimensions. If they keep their Atoms crappy, fine, everyone will buy ARM --- Intel has nothing special in their Atoms that make them compelling unless you REALLY need x86 compatibility. But if they improve their Atoms, then those Atoms become good enough for many of the purposes that used to require a CPU at 4 or 5 times the cost.
As if that weren't bad enough, they're facing the same problem on the server side. It's not obvious yet, but in a year or two it will be. Once again ARM vendors are willing to sell the same thing as Intel at half the cost. The ONLY advantage Intel has is x86 ISA, and that's becoming less and less important by the month. Once again Intel wants to limit all its weirdly acronymed Enterprise features to the expensive Xeons, but ARM vendors will give you those same features as part of the package on their low-end server chips.
Few companies have handled this sort of transition gracefully. Apple has --- they were willing to let iPhone cannibalize iPod, and iPad to cannibalize MBA (if in fact this happens --- some claim it does, though I doubt it). But for Intel it means having to match ARM prices, which is likely impossible the way Intel is currently set up. Intel probably COULD become the dominant ARM vendor if they were willing to accept the ARM ISA. But they are committed to x86, and the cost that that imposes in design overhead and verification makes it economically impossible for them to compete.
It'll be very interesting to see how their pricing structures and product segmentation evolve over the next two years...
Kidster3001 - Wednesday, November 19, 2014 - link
Your arguments make sense if cost is the only concern. They fall apart if you have any desire at all for performance. Intel's price/performance ratios can't be beat in servers. Silvermont is also better perf/watt than current Qualcomm cores. The rest of the SOC (Baytrail) is not so great, but the CPU cores are very good.DesDizzy - Thursday, November 20, 2014 - link
Thanx for your excellent summary of strategic position.pugster - Tuesday, November 18, 2014 - link
I doubt that Intel wants to compete with AMD at this point. The x86 cpu market has only 2 players left and if Intel decides to squeeze AMD out of the market the Anti-trust police will come after Intel. In fact, Intel is not doing anything about AMD into the APU market and this market is going to be stagnant.Intel has severely miscalculated the mobile market early on and now is paying the price. Intel always been selling of ditching divisions that is not making money just because of short term reasons like getting rid of their StrongARM business years back. Even so, Intel is doing things half-assed in their mobile dept because they are making chips in 22nm when they can make it at 14nm. Intel should be thinking of putting their chips on cell phones where there is a bigger profit margin compared to tablets where they can't compete of the likes of allwinner.
przemo_li - Wednesday, November 19, 2014 - link
Mobility is not fad.Its bigger then whole Intel.
Now Intel want to be bigger, and mobility is out there for grabbing. They just need tools to do that.
jimjamjamie - Tuesday, November 18, 2014 - link
Whatever gets rid of these useless Atoms (as they are now) is fine by me.darkich - Tuesday, November 18, 2014 - link
Actually the Core M so far seems to be far worse than the Bay Trail Moorefield, considering the price and process.If Intel wants to compete in the ULP space, I think they need to get rid of the desktop legacy arhitecture actually. Shrinking, restraining and restricting the desktop architecture bellow 5W TDP is not the way to go and the Core M so far proves it.
The fact that comparatively super cheap Apple A8X shows similar performance and far better efficiency on a less advanced process is a major blow to the myth of Intel.
Drumsticks - Wednesday, November 19, 2014 - link
I don't think A8X is showing similar performance to Core M, although I admittedly haven't seen a chart. It's an interesting point though, would Anandtech consider trying to do some kind of perf/watt benchmark? It might not be super easy, but would be really good to see.domboy - Tuesday, November 18, 2014 - link
I don't think the Atom CPU is useless... the Asus T100 is a nice little device for example. Sure it's no speed demon, but it can do a lot of useful things just fine.just4U - Tuesday, November 18, 2014 - link
Some degree of separation should be in play with the development of these processor families. It's helped Intel in the past and in some ways its reasonable to assume that it's part of the reason why they have what they do today.danjw - Tuesday, November 18, 2014 - link
This is supposed to help their mobile unit? One of Intel's problems is the lack of a radio on their mobile SOCs. How does splitting the mobile CPU and RF units help with that?I am wondering if this is going to be used as an excuse when they bump the ship date again for the Broadwell desktop parts.
ptmmac - Tuesday, November 18, 2014 - link
I expect this to be a better way to hide how much money their mobile division is losing. Intel needs the next big thing to drive a new era of exponential growth. The loss of Denson scaling has destroyed Intel's ablity to make its own older parts obsolete. Every 4 years a new PC was a requirement before 2004. Now a new PC is a luxury, and Mobile replaces half of the PC's market place with good enough content consumption machines.Intel has to get photonics and a new replacement for CMOS up and runing pronto. Then they can afford to eat their own success story. The cannot afford to let someone else destroy the PC.
coburn_c - Tuesday, November 18, 2014 - link
My bet is they are killing Atom and moving the mobile team over to focus on low power core derivatives.coburn_c - Tuesday, November 18, 2014 - link
To further clarify, they OoO'd the Atom and double the core count and it's still not competitive in performance. The Core is scaling down in power surprisingly well. Makes sense to kill the Atom.azazel1024 - Tuesday, November 18, 2014 - link
It is, but it still isn't competitive with Atom in one area where it really counts for mobile.Power draw. Core M is still higher power draw than Bay Trail-T (if not by huge amounts), its performance is not massively better (it IS better) and it deffinitely cannot be made in to a phone chip, which is part of where Intel is trying to go with Atom.
Now I could possibly see Intel refocusing Atom to be a phone only chip and do derivative Core designs to fill the tablet space from bottom to top, but I don't see them managing to make Core in to a phone chip ever. Laws of scaling, one design can really only do well with a 10x power factor. For Core, that means roughly 10-13w up to 100-130w. I'd semi-argue what Intel is really targeting is 9w-90w with the big iron really being the same architecture scaled out to extra cores. Core M makes some fairly big sacrafices to fit in to 5w, and looking at it, it appears to sacrafice performance per watt to get there. It most deffinitely does not seem to be an ideal design for that low a power consumption.
What Intel kind of needs is an even better Atom architecture that scales better from 1-10w, which I think is the direction they are moving. It is just going to take them a couple of years to get there. Airmont/Cherry Trail seems to be promising about a 15-30% increase in CPU performance (baring any unannounced major Architecture tweaks) and a 2-4x increase in GPU performance over Bay Trail. That'll GREATLY close the gap with Core M as it stands today. In fact it should blur almost all distinctions as it would likely leave Core M with only a small GPU advantage and possibly even a DISADVANTAGE in highly multithreaded applications, depending on throttling, workloads, etc.
Likely the next release of Atom, Goldmont/Willow Trail, is going to focus pretty heavily on new architecture for the CPU and might well yield some very large CPU gains.
That said, IMHO, Bay Trail makes an excellent tablet chip (I have a T100), a good netbook chip and a poor laptop chip (at least Bay Trail-T, but frankly Bay Trail-N/D isn't really that much better performing). Core M seems to make an okay laptop chip, an okay netbook chip and an okay tablet chip. In a couple more years, if Intel can keep iterating and delivering, Atom is likely to continue to make an excellent tablet chip and netbook chip and might even make an okay laptop chip.
savagemike - Tuesday, November 18, 2014 - link
Is Airmont still supposed to be released this year?I think everyone is over-reacting slightly to this management news. This kind of engineering takes time. Sharing management might impact designs on either the mobile or core front, but even if it does it won't show up in selling product for a couple years at least.
This is probably more an acknowledgement that at a certain point the hardware plateaus and there will be zero difference between the mobile parts and the low end desktop parts. So why have two divisions trying to design/build them? You'll just have one class of chip with the power efficiency to run mobile devices and the power capacity to service anything short of a work station. We're clearly on the doorstep of that era now.
DanNeely - Tuesday, November 18, 2014 - link
The process Intel uses for it's server, desktop and conventional laptop processors is different than the one they use for atom laptop chips and their attempted push into mobile. The latter's tuned for lower TDP instead of higher clock rates. The way Intel's pushing on ultrabooks though; assuming they can get the low power process caught up timeline wise, I wouldn't be surprised if they move ULP Core processors over to the latter too.przemo_li - Wednesday, November 19, 2014 - link
It may be strange to say, but it seam that Intel is blocked by GPU designs more then CPUs. Like this new Nokia tablet with Intel CPU and... GPU from somebody else.Either Intel price GPUs without those big incentives CPUs receive or Intel is not ready there.
From Linux work for next gen hw (hw AFTER broadwell), we can see that Intel is pursuing tilling GPUs, so they are aware of market requirements at least. Hope they are quick about it. They have some very good strong points beside performance (fully opensource cpu/gpu drivers!!!)
Kidster3001 - Wednesday, November 19, 2014 - link
Baytrail is the only Intel Atom ever to have in Intel GPU. All other Atoms have had 3rd party GPU's in them, PowerVR to be precise, the same GPU Apple uses in their Ax chips. The Nokia tablet uses Moorefield which has the PowerVR G6430 GPU in it.Krysto - Wednesday, November 19, 2014 - link
You're forgetting one thing. Intel's Core chips are 5-10x more expensive than Atoms. And Intel is already heavily subsidizing Atom. Core will be hilariously noncompetitive in the mobile market.