This is Intel's first full quarter with Ivy Bridge-EX out on the market. A lot of companies have been waiting for this chip as it brings a new platform (remember there was no Sandy Bridge-EX). I know of companies who are still using Nehalem-EX servers in production and looking to replace them after 4 years of good service.
Enterprise storage and networking are also doing pretty well for Intel.
The future is bright in this area with some revolutionary technologies like silicon photonics on the horizon.
The only real enemy in this space isn't a hardware competitor but rather crazy software licensing as socket, core and memory count increases. The hardware is good but if companies can't afford to run their software on it, demand will flat line over the long term. I expect Intel to become more involved in some open source projects as a message to several key enterprise developers.
They bought McAffee because they wanted/needed software to go with new security features in hardware. The problem is that they bought McAffee and not someone who could innovate. Software that went hand in hand for managing encryption on their SSDs, took advantage of their AES instructions to encrypt drives that are missing hardware based encryption. They should have bought Symantec. I don't think Symantec's software is great, but they do execute.and their software is decent. Intel should have bought Nvidia when AMD bought ATI. They've wasted a lot of money developing GPU hardware and software assets.
The problem with buying Symantec and Nvidia is the cost. Actually, buying Nvidia now makes more sense than ever before. They have a better mobile business than Intel (chips and modems), obviously GPUs, enterprise GPUs, CUDA, and software assets. And they had a great chipset business till Intel and AMD forced them out. Intel made a huge mistake by not letting Nvidia get an x86 development license; they forced them into the ARM camp.
Uh... Intel can't buy nvidia, as the purchase would be shot down by antitrust authorities. It's kind of hard to take the rest of your comment seriously when you don't even realize that much.
"All the cloud services are really helping them sell data center stuff"
It remains to be seen how sustainable this is. Intel have had no serious competition in the data center for years. POWER and Oracle/Sun are specialized (and expensive) tastes, AMD burns too much power, ARM has been limited to 32 bit. Over the next year the ARM story changes, with 64-bit available, the LAMP stack becomes standardized thanks to Linaro, and the ARM server spec becoming standardized. A57 should be, for most purposes good enough in latency (ie single-threaded throughput) but with the ability to run mass throughput at lower power and substantially lower cost than Xeon.
Intel may try to fight back by dramatically lowering the cost of Xeons, but while that may stave off ARM in the datacenter, it won't help sustain their revenue.
Data centers want higher IPC. They don't want to deploy 10x as many servers, even if the overall TDP of ARM is better, because there are large costs associated with housing and maintaining each and every server. As long as Intel leads in IPC they'll never need to lower Xeon prices.
More than 50% of costs for data centers come from POWER usage.
So no, neither IPC, nor power usage counts.
Power efficiency (how much we can do with given power budget) matter.
And if ARM SoCs offer improvements, without disruption, then they will be choosen over Intel offering in the blink of the eye.
Only saving grace Intel have currently is software support. And not "end-user software" as that is gcc recompile away (as it should be!), but firmware/drivers mess current "embeded" ARM OEMs create.
You can grab Intel source code, heck You do not even have to. Just Linux kernel (Intel code is included after all), compile it, slap into Your distro of choice, and You are ready.
With ARM You need those release-once-forget-about-customers-after-that binary blobs.
AMD went so far as to OpenSource their gaming GPU business for the sake of servers. ARM crowd laggs behind seriously.
Lack of flexibility -> higher costs of maintaining & less innovation.
I wonder if other companies besides Microsoft will start looking into FPGAs to get several times faster performance. Microsoft claimed to get 10x better performance for their FPGA based Catapult project
So mobile revenue of about 50mil but losing a billion on it. So basically you are completely GIVING these away to sell them...LOL. Quit and your earnings go up a full billion dollars.
Why does anandtech tell revenue instead of EARNINGS from each dept? Or heck better yet post BOTH. Intel paying you to NOT say how much mobile is losing (950mil last quarter)? Otherwise why leave out that VERY important detail to shareholders who say they should just give it up (many analysts say this). Not expected to put out anything decent on mobile until 2016 really and the competition won't sit still. We keep hearing, the next rev will do it...ROFL. Now they've given up that and say two revs it will happen (or more depending on who you believe).
By NV or give up mobile. Now that would be a HUGE game changer especially with Intel's process. Can you imagine qcom trying to keep up with a 14nm K1 Denver? Intel wouldn't have to GIVE these away, they'd sell themselves. Pay whatever Jen Hsun needs to leave or merge (a few billion on top of the buy price in a personal check should do it) and be done with it. Imagine the 14nm gpus coming out too. Again total dominance in mobile and gpu for a long time if they pump out 14nm NV stuff while owning them. Also I'm sure they'll be able to sue qcom at some point as IMHO they hide their gpu tech from everyone because they are stealing tech they should be paying for (or why hide while everyone else shows the cards?). Nobody is really using fancy tricks these days (which is why amd/nv basically tie each gen now), so what is qcom hiding? Intel would have the money and lawyers to take qcom down a notch at some point.
NV is the only way to stop the ARM juggernaut (if you can't beat them join then right?). Make a move Intel before arm enters desktops and takes 21% of that just like they did notebooks already. With android L and 64bit they can move to desktop for at least a decent portion of users who don't do much more than browse, email and casual gaming. That describes a LOT of users.
I don't believe the bump for intel will last long, it mainly happened because of xp dying via MS support. Companies were forced to look forward more.
I'm...not sure you know how to read financial statements. Also, Anand has no obligation to post any of this - he provided what was relevant - if you want to conduct a witch hunt at least link to the PUBLIC 10-Q which has all the info you're claiming.
On your last point, you're insane if you think antitrust lawyers would let INTC buy NVDA. If they would ever let it happen if would have happened in '07-'08 when INTC figured out Larrabee wasn't going to work.
The chart of process vs time is showing a pretty clear exponential decay curve. Barring a miracle, I'd say we're looking at an even longer time between 14 and 10nm.
Not really... you're forgetting that the gains from die shrinks are exponential. Going from 180nm to 130nm got us 1.9x as many transistors per amount of space, while going from 32nm to 22nm got us 2.1x as many transistors per amount of space.
It's the length of time that determines if things are slowing down, and while they did shift, there wasn't really any major change from one period to the next. Some were 7 quarters, some were 10 quarters, but 90nm and 32nm both lasted roughly similar amounts of time (only one quarter difference) and they're the better part of a decade apart. 14nm is the first one to really be slower, so while that might indicate that they're hitting a wall (and that 10nm will also be long in the making), the graph alone doesn't show that.
Indeed. One factor not mentioned is the desire to go to 450 mm wafers in that time frame. Producing more per wafer is going to be necessary with the need for more production time due to double/triple pattering.
The long 22 nm and 14 nm life spans to allow Intel to recover more on the capital investments into those lines. While the tick-tock time line looks like an exponential decay curve, the price of the fabs looks like an exponential growth curve.
As for 10 nm, it may be the last mass production silicon node. Intel is doing 7 nm prototyping and 5 nm research, but they're unsure they can bring them out of the lab in a cost effective manner. Even if they can get 7 nm to work on a mass production line, as I alluded too it may simply be too expensive to do.
What's hammering Intel and forcing a delay in 14nm is the same thing that's resulted in everyone else's 20nm processes being severely delayed (Intel's 22nm was just large enough not to need it). Scaling double patterning from the lab to production has proved much harder than anyone expected and blown everyone's timeline for their next node. This is, hopefully, a one off hit and the Intel's transition to 10nm along with all the half node company's transitions to 12(?)nm will return to the expected 2 year time scale.
Mobile accelerated their loss now to 1.1B. So you're giving away 1.15Billion in chips to make 50mil in revenue. How hard is it exactly to sell an Intel SOC? Over a billion dollars hard I guess...LOL. Buy NV or lose 1.5B per quarter by xmas. AT some point this hurts fab R&D. I mean we're talking 4B+ this year in losses on mobile, and it was almost that over that 12 months already. You're almost losing the cost of a whole fab...ROFL. Claiming you're on track to hit 40mil socs sold for tablets is a joke when you give them all away.
http://www.desktopreview.com/default.asp?newsID=13... $5B was the expected price of AZ fab that is currently empty in my state ;) Go figure. Open the fab and fab MORE stuff for others or eventually lose your butt to IBM/Samsung/GF (not so much GF but they gain also as IBM develops the tech and sammy/gf run with it, IBM just announced 3B for 7nm and below tech) as ARM cannibalizes some desktop sales (and at some point some server probably too) and yet more notebooks as they amp up perf with 64bit, 20nm that make them OK for low-end desktops at least. Surely at some point NV puts an NV card into an ARM desktop and really competes using OpenGL games pushed by google, valve etc. Surely a steamos port is being worked on for arm by now with close ties of valve/NV on steamos.
MS laying off now soon (thousands) and as soon as ARM marches on desktops I'd expect more as less WINTEL will be needed. Also russia going internal for arm chips, 8 core this year 16 core next year and again NO WINTEL. 700K PC's and 300K servers will no longer be bought every year from WINTEL. Layoffs at MS and an empty fab an Intel while mobile loses over a billion a quarter now. See that writing on the wall? It's yelling at you. Germany going linux too and at some point arm surely along with others due to NSA crap no doubt. NSA crap is really going to hurt usa companies over time (russia chips expected to hit routers & switches at some point to, to avoid spying), and they never should have went as far as they did for no real returns on tapping etc. Under oath they said how many crimes were stopped by all our billions of wasted dollars? ;) Now it's starting to cost our companies big time. Uh oh, I think I hear NSA knocking at my door...LOL.
It is true, Intel isn't as agile as it needs to be with SoC variety. They seemingly want to produce one or two jack of all trade SoC's instead of tailoring to the demands of specific OEMs for what they want to do for a specific end product. With Intel opening up their fabs and even licensing their cores to be part of 3rd party SoC's (see Rockchip) this appears to be changing.
However, if you are expecting ARM to sweep into the desktop is similar to the promise of Linux on the desktop. They're not going to happen for ultimately the exact same reason: legacy software. That means Windows running on x86 hardware. As demonstrated by many still hanging on to Windows XP after support has ended is an example of just how entrenched Windows + x86 is in the market. Sure, there is demand to move away from it due to MS licensing costs and the price of PC hardware vs. cheaper ARM SoC but competing products are worthless if they don't run the software you need.
The server market is a bit different. While x86 is dominate, that market has broken its dependance on Windows and x86 due to the fall out form the Unix wars a decade ago. Enterprise applications are commonly either open source and/or Java base to shed themselves of OS and hardware lock-in. As such ARM is expected to quickly move into the web server niche. However, the enterprise market values RAS and security, something ARM as a platform has yet to demonstrate in production.
I mostly agree with you. I do have one comment though: I think that the reason XP is still in use is because people don't care. There's no fundamental advantage using Windows7, except the "artificial" one imposed by Microsoft which is support. Obviously, Microsoft makes money selling Windows, not giving "free" support, so it is in their interest to push out new and "better" versions of Windows to push consumers to upgrade. But the truth is, up until XP, all previous Windows versions were "crap". XP was usable and intuitive. Windows7 is, arguably, better, but if you still had support for it, would you really be compelled to update? A Core2Duo still does a good job for a non-gaming PC, so why update? That's why they tried to "change the game" with Windows8 (and failed miserably). Now they're already on Windows9. Honestly, who in the world feels that Windows7 is inadequate? It's an OS for crying out loud: I use software, not the OS. As long as it's stable, I'm happy.
You are conflating a whole lot of separate markets as a single "desktop" market.
There are special-purpose devices (like ATMs, cash registers, industrial control machines). No-one cares about these --- they can run off the cheapest Atom, and they run until they die. They're not going to keep Intel rich.
There are ENTERPRISE desktops. These think they need to run Office and some PoS customized app written in Visual Basic in 1997 and never updated since then. THAT is when Intel and MS have a lock.
There are HOME desktops (to a large extent this also includes schools and colleges and small businesses). These are much more flexible in what they run, and don't care about the details as long as they get functionality -- they want word processing, not Word, or Browsing, not IE. We are already seeing this space fragment. This is where Apple, of course, lives. But also Chromebooks (to the extent they are sold) and plenty of iOS and Android activity in varying form factors. All this activity makes Wintel compatibility just not that valuable.
So basically Wintel has one lockin, the enterprise. This is a valuable lockin, but it's not massively growing; on the contrary it gets nibbled at every day. The Apple/IBM alliance announced yesterday is one more part in this nibbling. It's a small part (for now) but sets the tone.
Part of MS' problem is that we now have a pretty good model, which covers a large number of use cases, for how to replace that software written in 1998 --- have the backend serve up HTML which gets displayed either in a web browser or some sort of mobile wrapper app. I expect many of the IBM replacement mobile apps they talk about to be written that way, and every one that does is a further proof of concept. This doesn't help a few specialized cases (people who push Excel hard, a very limited market; people who use Office VB macros to design an enterprise workflow, a rather larger market) but it means Wintel SW lockin is fighting an ongoing battle, and losing a little every month.
ARM marches on desktops - what a joke, the fastest ARM chips barely outperform an Atom and no-one is suggesting putting them in a desktop. It is more likely Intel eats Arm at the higher end of mobile with Broadwell than Arm eats Intel in desktop.
Look at what the market is telling you, since March Intel is up 23% while ARM is down 19%.
Giving away 40m tablet chips is the right strategy, once people try a Wintel tablet, they are not going to go back to some crappy android tablet. Even my iPad Air is barely used for anything other than surfing the web or watching movies. Trying to do anything productive on it is like slamming your head against a wall.
"ARM marches on desktops - what a joke, the fastest ARM chips barely outperform an Atom and no-one is suggesting putting them in a desktop."
The fastest ARM chip is Apple A7. This performs equivalent to an i3/i5/i7 at equivalent GHz. iPad Air has an A7 at 1.4GHz. MBA has an i5 at 1.6GHz. MBA feels faster because the i5 turbos up to 2.4GHz, and Intel turbo'ing is so good that it can run most "snappy" UI stuff at the 2.4GHz. There is every reason to expect that A8 will run at same IPC but clocked at 2GHz. A6 was a 2x speedboost over A5. (about 1.6 from process, 1.2 from micro-architecture) A7 was a 1.5x speedboost, all micro-architecture. I'd expect A8 to be about a 1.5x speedboost, about 1.4x from process and 1.1 from micro-architecture). In the six years since Nehalem, Intel has sped up by about 1.25x, all microarchitecture since all their process has been used in lowering power. (This is not quite true, and is especially misleading in the Y category, because turboing has been used to give substantial frequency boosts to nominally low frequency ULV processors).
OK, that's the state of the art today. Extrapolating to the future, the patterns that matter are - Apple is vastly more agile than Intel. This is primarily because they don't have the baggage of 30+ years of accumulated poor ideas that have been added to the x86, none of which have ever been removed.
- Apple has had it easy so far, especially in improving micro-architecture, because they have been playing catchup. There is not much scope for improvement there as long as they stick to conventional micro-architecture. BUT there is also no reason that they have to stick to conventional micro-architecture. They aren't Intel, terrified of change and with an unbelievable test burden. They are in a better position than anyone else to experiment with an unconventional micro-architecture like kilo-insruction processing, which could net them another 30% or so.
- Apple have managed to match Intel with a core that is smaller (100mm^2 for the ENTIRE A7, about 35% of which is stuff that's not CPU or GPU), cheaper to make, on a lower process. The ARM hoard will catch up to this point eventually. They can ALSO move faster than Apple, and they will be more aggressive given that Apple has shown proof of concept.
- There seems no intrinsic reason why Apple cannot, over the next three or four years, copy the parts of Intel that matter and which they are missing to reach parity. More cores is trivial. Higher frequency is primarily a question of how much power you're willing to burn. If Apple wanted, they could copy AVX2 and go with wider vectors, but I think they see HSA as more promising way to solve that problem. Better uncore (NoC, common NUCA-based L3, etc) are coming soon, I expect in the A8. HW transactional memory is coming soon, I also expect in the A8.
The issue is NOT that Intel is ahead today. The issue is that - Intel is less agile - Intel has no compelling technical advantages - Intel cannot survive on the cost structure that the ARM ecosystem is used to.
Intel won't be replaced on the desktop tomorrow. But in five years...?
Not surprising in the least, I just realized a few days ago that I had purchased 4 Haswell-based products over the last year (2x4770K desktop builds, 1xNUC, 1xSurface Pro 3). Intel really did an amazing job scaling this architecture for all work loads while maintaining high performance and full x86 Windows compatibility.
I think we will see more of this in the future and this was the first quarter where the market came to the same realization that I did over a year ago. I'm done with the novelty of all these gadgets running toy OSes like Android and iOS, when it comes down to business, gaming and productivity, there's just no replacing a full blown x86 ISA chip running full blown Windows. As much as Win8/.1 are maligned, it's really a better OS after you get past the fullscreen Modern UI (which is thankfully going away with Threshold).
To be fair, there's no reason Windows can't run as well on an ARM chip. There's just the matter of older software, but anything new could be cross-compiled if Microsoft allowed binary executables on the Win8 desktop.
I completely agree with you on Windows, though. I simply couldn't have a productivity machine that uses Android or iOS. People have debated between the interfaces between OSX and Windows for years; but when you bring the argument to things like Android, Windows wins hands-down. It's just a lot easier to use.
According to everybody, Intel included, the refresh in business PCs caused by XP's EOL is driving sales but consumer is still soft. So not much of a good news since it's a temporary thing and lets face it ,the worst is still to come for them. For now other devices have made the traditional PC less relevant but those new devices will start to replace PCs at some point - it should have started already but the mobile players don't seem too interested. 14nm is in theory arriving late this year but it seems very low volume and a proper roll out will be over the first half of next year. 10nm might have to come rather fast or the foundries might catch up, things are heating up there and chances are TSMC and Samsung will be racing each other hard. Hope AMD's new cores will be competitive because this lack of competition is just allowing Intel to offer less and less, else at some point soon they'll start selling us mobile SoCs sized chips at 300$, just because they can.
Do you mean sized or powered? As it is, baytrail is ~100mm2. Haswell is north of 175mm2. As gfx becomes more and more prominent that gap is going to narrow, especially as Atom and Core start using the same gfx subsystems. As for power, that will also narrow a lot. My speculation is that fanless Broadwell and Skylake will maybe, maybe be 25% faster on the CPU side than Cherrytrail/Broxton. That's simply due to a TDP limitation.
ARM isn't a danger in desktops because desktops are slowly being marginalized. The only desktop like comps that are worth anything will be workstation grade stuff when you need the heavy lifting a Intel chip will give you.
Intel is in no danger of becoming irrelevant as worst thing that could happen is they become the worlds most profitable fab company even if TSMC or Samsung catch up. As long as the barrier to entry for foundries is approaching $10b they won't have a level of competition to eat away their revenues for a while.
Long term I'm also not so sure how profitable mobile chips will be as a segment. The value is all in the fab and chip design is becoming commoditized with the fabless semi companies. When 3g is no longer necessary for modems Qualcomm loses their biggest leverage and their margins will drop.
Intel will need to worry about the server market which has been and will be their cash cow, mobile will be a plus but margins will fall and hurt the stock if they try and compete to heavily in mobile.
If Otenelli had taken the margin hit by fully committing to mobile first before he retired things might be different but that ship has sailed and the new CEO may not have the job security necessary to lose that much money by foregoing xeons for mobile.
Remember AMD hot killed because everyone expected them to make Intel sized margins and they just couldnt because Intel was the outlier and only in retrospect do people realize that AMD tossed their most important asset by splitting with GloFo. The most valuable asset AMD has left is their cross licensing deal with Intel from the 64 bit transition that allows them to use Intel x86 IP and kept the feds from opening an antitrust case against Intel.
The future is unwritten but Intel will be fine but not the dominant player it once was.
Not really...technology inherently favors monopolies with high margins (see also: Apple in tablets/phones, Google and Facebook in their areas). Those high margins are necessary to feed the RnD and Capital spending Intel needs to drive new processes. Net Margin is only 20%, so 2/3 of the gross gets eaten up feeding new technologies and growth.
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
33 Comments
Back to Article
skiboysteve - Tuesday, July 15, 2014 - link
1) wow what a great quarter for Intel! All the cloud services are really helping them sell data center stuff2) why the hell did they buy McAfee?
Kevin G - Wednesday, July 16, 2014 - link
This is Intel's first full quarter with Ivy Bridge-EX out on the market. A lot of companies have been waiting for this chip as it brings a new platform (remember there was no Sandy Bridge-EX). I know of companies who are still using Nehalem-EX servers in production and looking to replace them after 4 years of good service.Enterprise storage and networking are also doing pretty well for Intel.
The future is bright in this area with some revolutionary technologies like silicon photonics on the horizon.
The only real enemy in this space isn't a hardware competitor but rather crazy software licensing as socket, core and memory count increases. The hardware is good but if companies can't afford to run their software on it, demand will flat line over the long term. I expect Intel to become more involved in some open source projects as a message to several key enterprise developers.
eanazag - Wednesday, July 16, 2014 - link
They bought McAffee because they wanted/needed software to go with new security features in hardware. The problem is that they bought McAffee and not someone who could innovate. Software that went hand in hand for managing encryption on their SSDs, took advantage of their AES instructions to encrypt drives that are missing hardware based encryption. They should have bought Symantec. I don't think Symantec's software is great, but they do execute.and their software is decent. Intel should have bought Nvidia when AMD bought ATI. They've wasted a lot of money developing GPU hardware and software assets.The problem with buying Symantec and Nvidia is the cost. Actually, buying Nvidia now makes more sense than ever before. They have a better mobile business than Intel (chips and modems), obviously GPUs, enterprise GPUs, CUDA, and software assets. And they had a great chipset business till Intel and AMD forced them out. Intel made a huge mistake by not letting Nvidia get an x86 development license; they forced them into the ARM camp.
sonicmerlin - Wednesday, July 16, 2014 - link
Uh... Intel can't buy nvidia, as the purchase would be shot down by antitrust authorities. It's kind of hard to take the rest of your comment seriously when you don't even realize that much.name99 - Wednesday, July 16, 2014 - link
"All the cloud services are really helping them sell data center stuff"It remains to be seen how sustainable this is. Intel have had no serious competition in the data center for years. POWER and Oracle/Sun are specialized (and expensive) tastes, AMD burns too much power, ARM has been limited to 32 bit.
Over the next year the ARM story changes, with 64-bit available, the LAMP stack becomes standardized thanks to Linaro, and the ARM server spec becoming standardized. A57 should be, for most purposes good enough in latency (ie single-threaded throughput) but with the ability to run mass throughput at lower power and substantially lower cost than Xeon.
Intel may try to fight back by dramatically lowering the cost of Xeons, but while that may stave off ARM in the datacenter, it won't help sustain their revenue.
sonicmerlin - Wednesday, July 16, 2014 - link
Data centers want higher IPC. They don't want to deploy 10x as many servers, even if the overall TDP of ARM is better, because there are large costs associated with housing and maintaining each and every server. As long as Intel leads in IPC they'll never need to lower Xeon prices.przemo_li - Monday, July 21, 2014 - link
More than 50% of costs for data centers come from POWER usage.So no, neither IPC, nor power usage counts.
Power efficiency (how much we can do with given power budget) matter.
And if ARM SoCs offer improvements, without disruption, then they will be choosen over Intel offering in the blink of the eye.
Only saving grace Intel have currently is software support. And not "end-user software" as that is gcc recompile away (as it should be!), but firmware/drivers mess current "embeded" ARM OEMs create.
You can grab Intel source code, heck You do not even have to. Just Linux kernel (Intel code is included after all), compile it, slap into Your distro of choice, and You are ready.
With ARM You need those release-once-forget-about-customers-after-that binary blobs.
AMD went so far as to OpenSource their gaming GPU business for the sake of servers. ARM crowd laggs behind seriously.
Lack of flexibility -> higher costs of maintaining & less innovation.
jobo32 - Thursday, July 17, 2014 - link
I wonder if other companies besides Microsoft will start looking into FPGAs to get several times faster performance. Microsoft claimed to get 10x better performance for their FPGA based Catapult projectprzemo_li - Monday, July 21, 2014 - link
Intel is integrating some FPGA's in their server lineup.So everybody will be able to try their luck with them.
TheJian - Tuesday, July 15, 2014 - link
So mobile revenue of about 50mil but losing a billion on it. So basically you are completely GIVING these away to sell them...LOL. Quit and your earnings go up a full billion dollars.Why does anandtech tell revenue instead of EARNINGS from each dept? Or heck better yet post BOTH. Intel paying you to NOT say how much mobile is losing (950mil last quarter)? Otherwise why leave out that VERY important detail to shareholders who say they should just give it up (many analysts say this). Not expected to put out anything decent on mobile until 2016 really and the competition won't sit still. We keep hearing, the next rev will do it...ROFL. Now they've given up that and say two revs it will happen (or more depending on who you believe).
By NV or give up mobile. Now that would be a HUGE game changer especially with Intel's process. Can you imagine qcom trying to keep up with a 14nm K1 Denver? Intel wouldn't have to GIVE these away, they'd sell themselves. Pay whatever Jen Hsun needs to leave or merge (a few billion on top of the buy price in a personal check should do it) and be done with it. Imagine the 14nm gpus coming out too. Again total dominance in mobile and gpu for a long time if they pump out 14nm NV stuff while owning them. Also I'm sure they'll be able to sue qcom at some point as IMHO they hide their gpu tech from everyone because they are stealing tech they should be paying for (or why hide while everyone else shows the cards?). Nobody is really using fancy tricks these days (which is why amd/nv basically tie each gen now), so what is qcom hiding? Intel would have the money and lawyers to take qcom down a notch at some point.
NV is the only way to stop the ARM juggernaut (if you can't beat them join then right?). Make a move Intel before arm enters desktops and takes 21% of that just like they did notebooks already. With android L and 64bit they can move to desktop for at least a decent portion of users who don't do much more than browse, email and casual gaming. That describes a LOT of users.
I don't believe the bump for intel will last long, it mainly happened because of xp dying via MS support. Companies were forced to look forward more.
ai744 - Tuesday, July 15, 2014 - link
Quite the load of delusional nonsense you’ve got here.stadisticado - Wednesday, July 16, 2014 - link
I'm...not sure you know how to read financial statements. Also, Anand has no obligation to post any of this - he provided what was relevant - if you want to conduct a witch hunt at least link to the PUBLIC 10-Q which has all the info you're claiming.On your last point, you're insane if you think antitrust lawyers would let INTC buy NVDA. If they would ever let it happen if would have happened in '07-'08 when INTC figured out Larrabee wasn't going to work.
A5 - Tuesday, July 15, 2014 - link
The chart of process vs time is showing a pretty clear exponential decay curve. Barring a miracle, I'd say we're looking at an even longer time between 14 and 10nm.Guspaz - Tuesday, July 15, 2014 - link
Not really... you're forgetting that the gains from die shrinks are exponential. Going from 180nm to 130nm got us 1.9x as many transistors per amount of space, while going from 32nm to 22nm got us 2.1x as many transistors per amount of space.It's the length of time that determines if things are slowing down, and while they did shift, there wasn't really any major change from one period to the next. Some were 7 quarters, some were 10 quarters, but 90nm and 32nm both lasted roughly similar amounts of time (only one quarter difference) and they're the better part of a decade apart. 14nm is the first one to really be slower, so while that might indicate that they're hitting a wall (and that 10nm will also be long in the making), the graph alone doesn't show that.
Kevin G - Wednesday, July 16, 2014 - link
Indeed. One factor not mentioned is the desire to go to 450 mm wafers in that time frame. Producing more per wafer is going to be necessary with the need for more production time due to double/triple pattering.The long 22 nm and 14 nm life spans to allow Intel to recover more on the capital investments into those lines. While the tick-tock time line looks like an exponential decay curve, the price of the fabs looks like an exponential growth curve.
As for 10 nm, it may be the last mass production silicon node. Intel is doing 7 nm prototyping and 5 nm research, but they're unsure they can bring them out of the lab in a cost effective manner. Even if they can get 7 nm to work on a mass production line, as I alluded too it may simply be too expensive to do.
DanNeely - Wednesday, July 16, 2014 - link
What's hammering Intel and forcing a delay in 14nm is the same thing that's resulted in everyone else's 20nm processes being severely delayed (Intel's 22nm was just large enough not to need it). Scaling double patterning from the lab to production has proved much harder than anyone expected and blown everyone's timeline for their next node. This is, hopefully, a one off hit and the Intel's transition to 10nm along with all the half node company's transitions to 12(?)nm will return to the expected 2 year time scale.TheJian - Tuesday, July 15, 2014 - link
http://news.investors.com/technology/071514-708941...Mobile accelerated their loss now to 1.1B. So you're giving away 1.15Billion in chips to make 50mil in revenue. How hard is it exactly to sell an Intel SOC? Over a billion dollars hard I guess...LOL. Buy NV or lose 1.5B per quarter by xmas. AT some point this hurts fab R&D. I mean we're talking 4B+ this year in losses on mobile, and it was almost that over that 12 months already. You're almost losing the cost of a whole fab...ROFL. Claiming you're on track to hit 40mil socs sold for tablets is a joke when you give them all away.
http://www.desktopreview.com/default.asp?newsID=13...
$5B was the expected price of AZ fab that is currently empty in my state ;) Go figure. Open the fab and fab MORE stuff for others or eventually lose your butt to IBM/Samsung/GF (not so much GF but they gain also as IBM develops the tech and sammy/gf run with it, IBM just announced 3B for 7nm and below tech) as ARM cannibalizes some desktop sales (and at some point some server probably too) and yet more notebooks as they amp up perf with 64bit, 20nm that make them OK for low-end desktops at least. Surely at some point NV puts an NV card into an ARM desktop and really competes using OpenGL games pushed by google, valve etc. Surely a steamos port is being worked on for arm by now with close ties of valve/NV on steamos.
MS laying off now soon (thousands) and as soon as ARM marches on desktops I'd expect more as less WINTEL will be needed. Also russia going internal for arm chips, 8 core this year 16 core next year and again NO WINTEL. 700K PC's and 300K servers will no longer be bought every year from WINTEL. Layoffs at MS and an empty fab an Intel while mobile loses over a billion a quarter now. See that writing on the wall? It's yelling at you. Germany going linux too and at some point arm surely along with others due to NSA crap no doubt. NSA crap is really going to hurt usa companies over time (russia chips expected to hit routers & switches at some point to, to avoid spying), and they never should have went as far as they did for no real returns on tapping etc. Under oath they said how many crimes were stopped by all our billions of wasted dollars? ;) Now it's starting to cost our companies big time. Uh oh, I think I hear NSA knocking at my door...LOL.
A5 - Tuesday, July 15, 2014 - link
No one who mistrusts US equipment for NSA reasons is going to use Russian chips. Come on.Kevin G - Wednesday, July 16, 2014 - link
It is true, Intel isn't as agile as it needs to be with SoC variety. They seemingly want to produce one or two jack of all trade SoC's instead of tailoring to the demands of specific OEMs for what they want to do for a specific end product. With Intel opening up their fabs and even licensing their cores to be part of 3rd party SoC's (see Rockchip) this appears to be changing.However, if you are expecting ARM to sweep into the desktop is similar to the promise of Linux on the desktop. They're not going to happen for ultimately the exact same reason: legacy software. That means Windows running on x86 hardware. As demonstrated by many still hanging on to Windows XP after support has ended is an example of just how entrenched Windows + x86 is in the market. Sure, there is demand to move away from it due to MS licensing costs and the price of PC hardware vs. cheaper ARM SoC but competing products are worthless if they don't run the software you need.
The server market is a bit different. While x86 is dominate, that market has broken its dependance on Windows and x86 due to the fall out form the Unix wars a decade ago. Enterprise applications are commonly either open source and/or Java base to shed themselves of OS and hardware lock-in. As such ARM is expected to quickly move into the web server niche. However, the enterprise market values RAS and security, something ARM as a platform has yet to demonstrate in production.
yankeeDDL - Wednesday, July 16, 2014 - link
I mostly agree with you.I do have one comment though: I think that the reason XP is still in use is because people don't care.
There's no fundamental advantage using Windows7, except the "artificial" one imposed by Microsoft which is support.
Obviously, Microsoft makes money selling Windows, not giving "free" support, so it is in their interest to push out new and "better" versions of Windows to push consumers to upgrade.
But the truth is, up until XP, all previous Windows versions were "crap". XP was usable and intuitive. Windows7 is, arguably, better, but if you still had support for it, would you really be compelled to update? A Core2Duo still does a good job for a non-gaming PC, so why update?
That's why they tried to "change the game" with Windows8 (and failed miserably). Now they're already on Windows9. Honestly, who in the world feels that Windows7 is inadequate? It's an OS for crying out loud: I use software, not the OS. As long as it's stable, I'm happy.
name99 - Wednesday, July 16, 2014 - link
You are conflating a whole lot of separate markets as a single "desktop" market.There are special-purpose devices (like ATMs, cash registers, industrial control machines). No-one cares about these --- they can run off the cheapest Atom, and they run until they die. They're not going to keep Intel rich.
There are ENTERPRISE desktops. These think they need to run Office and some PoS customized app written in Visual Basic in 1997 and never updated since then. THAT is when Intel and MS have a lock.
There are HOME desktops (to a large extent this also includes schools and colleges and small businesses). These are much more flexible in what they run, and don't care about the details as long as they get functionality -- they want word processing, not Word, or Browsing, not IE. We are already seeing this space fragment. This is where Apple, of course, lives. But also Chromebooks (to the extent they are sold) and plenty of iOS and Android activity in varying form factors. All this activity makes Wintel compatibility just not that valuable.
So basically Wintel has one lockin, the enterprise. This is a valuable lockin, but it's not massively growing; on the contrary it gets nibbled at every day. The Apple/IBM alliance announced yesterday is one more part in this nibbling. It's a small part (for now) but sets the tone.
Part of MS' problem is that we now have a pretty good model, which covers a large number of use cases, for how to replace that software written in 1998 --- have the backend serve up HTML which gets displayed either in a web browser or some sort of mobile wrapper app. I expect many of the IBM replacement mobile apps they talk about to be written that way, and every one that does is a further proof of concept.
This doesn't help a few specialized cases (people who push Excel hard, a very limited market; people who use Office VB macros to design an enterprise workflow, a rather larger market) but it means Wintel SW lockin is fighting an ongoing battle, and losing a little every month.
Speedfriend - Wednesday, July 16, 2014 - link
@TheJianARM marches on desktops - what a joke, the fastest ARM chips barely outperform an Atom and no-one is suggesting putting them in a desktop. It is more likely Intel eats Arm at the higher end of mobile with Broadwell than Arm eats Intel in desktop.
Look at what the market is telling you, since March Intel is up 23% while ARM is down 19%.
Giving away 40m tablet chips is the right strategy, once people try a Wintel tablet, they are not going to go back to some crappy android tablet. Even my iPad Air is barely used for anything other than surfing the web or watching movies. Trying to do anything productive on it is like slamming your head against a wall.
name99 - Wednesday, July 16, 2014 - link
"ARM marches on desktops - what a joke, the fastest ARM chips barely outperform an Atom and no-one is suggesting putting them in a desktop."The fastest ARM chip is Apple A7. This performs equivalent to an i3/i5/i7 at equivalent GHz.
iPad Air has an A7 at 1.4GHz. MBA has an i5 at 1.6GHz. MBA feels faster because the i5 turbos up to 2.4GHz, and Intel turbo'ing is so good that it can run most "snappy" UI stuff at the 2.4GHz.
There is every reason to expect that A8 will run at same IPC but clocked at 2GHz.
A6 was a 2x speedboost over A5. (about 1.6 from process, 1.2 from micro-architecture)
A7 was a 1.5x speedboost, all micro-architecture. I'd expect A8 to be about a 1.5x speedboost, about 1.4x from process and 1.1 from micro-architecture).
In the six years since Nehalem, Intel has sped up by about 1.25x, all microarchitecture since all their process has been used in lowering power. (This is not quite true, and is especially misleading in the Y category, because turboing has been used to give substantial frequency boosts to nominally low frequency ULV processors).
OK, that's the state of the art today. Extrapolating to the future, the patterns that matter are
- Apple is vastly more agile than Intel. This is primarily because they don't have the baggage of 30+ years of accumulated poor ideas that have been added to the x86, none of which have ever been removed.
- Apple has had it easy so far, especially in improving micro-architecture, because they have been playing catchup. There is not much scope for improvement there as long as they stick to conventional micro-architecture. BUT there is also no reason that they have to stick to conventional micro-architecture. They aren't Intel, terrified of change and with an unbelievable test burden. They are in a better position than anyone else to experiment with an unconventional micro-architecture like kilo-insruction processing, which could net them another 30% or so.
- Apple have managed to match Intel with a core that is smaller (100mm^2 for the ENTIRE A7, about 35% of which is stuff that's not CPU or GPU), cheaper to make, on a lower process. The ARM hoard will catch up to this point eventually. They can ALSO move faster than Apple, and they will be more aggressive given that Apple has shown proof of concept.
- There seems no intrinsic reason why Apple cannot, over the next three or four years, copy the parts of Intel that matter and which they are missing to reach parity. More cores is trivial. Higher frequency is primarily a question of how much power you're willing to burn. If Apple wanted, they could copy AVX2 and go with wider vectors, but I think they see HSA as more promising way to solve that problem. Better uncore (NoC, common NUCA-based L3, etc) are coming soon, I expect in the A8. HW transactional memory is coming soon, I also expect in the A8.
The issue is NOT that Intel is ahead today. The issue is that
- Intel is less agile
- Intel has no compelling technical advantages
- Intel cannot survive on the cost structure that the ARM ecosystem is used to.
Intel won't be replaced on the desktop tomorrow. But in five years...?
Speedfriend - Thursday, July 17, 2014 - link
"The fastest ARM chip is Apple A7. This performs equivalent to an i3/i5/i7 at equivalent GHz."No, an i7 at 1.4ghz, using only 2 cores scores half the A7 time on sunspider.
There is also no evidence of the A7's ability to be clocked to anywhere near the level that the i5/i7 run at.
From all accounts, the A7 is way behind Haswell on a performance per watt basis.
There is far more evidence that Intel is moving down the power consumption curve towards Apple than Apple moving up the processor power curve.
chizow - Tuesday, July 15, 2014 - link
Not surprising in the least, I just realized a few days ago that I had purchased 4 Haswell-based products over the last year (2x4770K desktop builds, 1xNUC, 1xSurface Pro 3). Intel really did an amazing job scaling this architecture for all work loads while maintaining high performance and full x86 Windows compatibility.I think we will see more of this in the future and this was the first quarter where the market came to the same realization that I did over a year ago. I'm done with the novelty of all these gadgets running toy OSes like Android and iOS, when it comes down to business, gaming and productivity, there's just no replacing a full blown x86 ISA chip running full blown Windows. As much as Win8/.1 are maligned, it's really a better OS after you get past the fullscreen Modern UI (which is thankfully going away with Threshold).
mkozakewich - Wednesday, July 16, 2014 - link
To be fair, there's no reason Windows can't run as well on an ARM chip. There's just the matter of older software, but anything new could be cross-compiled if Microsoft allowed binary executables on the Win8 desktop.I completely agree with you on Windows, though. I simply couldn't have a productivity machine that uses Android or iOS. People have debated between the interfaces between OSX and Windows for years; but when you bring the argument to things like Android, Windows wins hands-down. It's just a lot easier to use.
jjj - Wednesday, July 16, 2014 - link
According to everybody, Intel included, the refresh in business PCs caused by XP's EOL is driving sales but consumer is still soft. So not much of a good news since it's a temporary thing and lets face it ,the worst is still to come for them. For now other devices have made the traditional PC less relevant but those new devices will start to replace PCs at some point - it should have started already but the mobile players don't seem too interested.14nm is in theory arriving late this year but it seems very low volume and a proper roll out will be over the first half of next year.
10nm might have to come rather fast or the foundries might catch up, things are heating up there and chances are TSMC and Samsung will be racing each other hard.
Hope AMD's new cores will be competitive because this lack of competition is just allowing Intel to offer less and less, else at some point soon they'll start selling us mobile SoCs sized chips at 300$, just because they can.
stadisticado - Wednesday, July 16, 2014 - link
Do you mean sized or powered? As it is, baytrail is ~100mm2. Haswell is north of 175mm2. As gfx becomes more and more prominent that gap is going to narrow, especially as Atom and Core start using the same gfx subsystems. As for power, that will also narrow a lot. My speculation is that fanless Broadwell and Skylake will maybe, maybe be 25% faster on the CPU side than Cherrytrail/Broxton. That's simply due to a TDP limitation.errorr - Wednesday, July 16, 2014 - link
ARM isn't a danger in desktops because desktops are slowly being marginalized. The only desktop like comps that are worth anything will be workstation grade stuff when you need the heavy lifting a Intel chip will give you.Intel is in no danger of becoming irrelevant as worst thing that could happen is they become the worlds most profitable fab company even if TSMC or Samsung catch up. As long as the barrier to entry for foundries is approaching $10b they won't have a level of competition to eat away their revenues for a while.
Long term I'm also not so sure how profitable mobile chips will be as a segment. The value is all in the fab and chip design is becoming commoditized with the fabless semi companies. When 3g is no longer necessary for modems Qualcomm loses their biggest leverage and their margins will drop.
Intel will need to worry about the server market which has been and will be their cash cow, mobile will be a plus but margins will fall and hurt the stock if they try and compete to heavily in mobile.
If Otenelli had taken the margin hit by fully committing to mobile first before he retired things might be different but that ship has sailed and the new CEO may not have the job security necessary to lose that much money by foregoing xeons for mobile.
Remember AMD hot killed because everyone expected them to make Intel sized margins and they just couldnt because Intel was the outlier and only in retrospect do people realize that AMD tossed their most important asset by splitting with GloFo. The most valuable asset AMD has left is their cross licensing deal with Intel from the 64 bit transition that allows them to use Intel x86 IP and kept the feds from opening an antitrust case against Intel.
The future is unwritten but Intel will be fine but not the dominant player it once was.
Hrel - Wednesday, July 16, 2014 - link
We need competition so badly... those margins are insane.stadisticado - Wednesday, July 16, 2014 - link
Not really...technology inherently favors monopolies with high margins (see also: Apple in tablets/phones, Google and Facebook in their areas). Those high margins are necessary to feed the RnD and Capital spending Intel needs to drive new processes. Net Margin is only 20%, so 2/3 of the gross gets eaten up feeding new technologies and growth.darkich - Thursday, July 17, 2014 - link
The 82% yearly drop in the already abysmal mobile division revenues is by far the most interesting part of this report.So I can conclude that the magnitude of Silvermont architecture fail just cannot be overstated.
darkich - Thursday, July 17, 2014 - link
2 BILLION LOSS in mobile, with their most ambitious mobile chip effort yet!http://www.extremetech.com/extreme/186367-intels-m...