Intel made a huge mistake 10 years ago
vox.com
vox.com
A more direct source, from Intel's own CEO (at the time): http://www.theinquirer.net/inquirer/news/2268985/outgoing-in...
Otellini said Intel passed on the opportunity to supply Apple because the economics did not make sense at the time given the forecast product cost and expected volume. He told The Atlantic, "The thing you have to remember is that this was before the iPhone was introduced and no one knew what the iPhone would do... At the end of the day, there was a chip that they were interested in that they wanted to pay a certain price for and not a nickel more and that price was below our forecasted cost. I couldn't see it.
"It wasn't one of these things you can make up on volume. And in hindsight, the forecasted cost was wrong and the volume was 100x what anyone thought."
Intel has made a bet that it would take just as much time for ARM to reach x86 level of performance as it would take Intel to bring x86 power consumption down to SOC levels. Now Intel can play on both sides with their low power x86 parts, Xeon-D with upto 16 cores and Core-M in 2c/4t configuration.
There's only one little problem: the market has settled on ARM and doesn't care anymore.
So how are they gonna play those "both sides" again?
The Intel Houdini software used to make Android on X86 possible is a marvel of engineering but it is not perfect. I worked for a company that made an Android tablet around an Intel chip instead of ARM. We were hit with a great number of user reports of certain apps crashing, overheating, or simply refusing to run in the first place. We verified in QA that these apps would work just fine on ARM tablets with the same Android version and similar specs.
To be fair to Intel, they were very good to work with and really did (and do) want to make it work. I have no doubt they will. Eventually.
Currently iOS is the only one that also supports having C and C++ as bytecode (LLVM one).
Android and WP, only support it for Java, RenderScript and .NET languages.
[0] https://cloudplatform.googleblog.com/2016/04/Google-and-Rack...
Second, they cost WAYYYY more than a mobile chip. They cost more than an iPhone's ENTIRE BOM, or about 10x more than what Apple pays for its chips.
Throttling has been a problem on high-end Androids for years now.
> Second, they cost WAYYYY more than a mobile chip.
That's the real issue – the margins in the mobile market are ridiculously slim, and there is no x86 monopoly Intel can abuse to lock in vendors and customers. So cheap ARMs it is.
Atom has barely increased its single-threaded performance from like 300 pts in Passmark to about 540 right now (in Pentiums, in PCs, which cost $160 a pop).
Intel is not competitive on price. That's why Atom didn't pan out in mobile. They tried to do it by heavily subsidizing its high-end mobile chips, which typically cost close to $50, to compete against a $15 mid-range Qualcomm chip. That's why some people were "impressed" by what an Asus Zenfone could do for instance, for its mid-range price overall.
And if Intel can't compete with Atom on price, then there's NO CHANCE it will ever compete with Core M in mobile. The only reason it even exists in so called "tablets" that go for $1000 right now, is because Microsoft failed to make a good case for ARM-based Windows machines with its poor app support. But that will always remain a niche.
And outside the Surface those aren't selling well either.
(IMO rightfully so, the Venue 11 Pro is the worst device I've seen in years – hypothetical hardware dickwaving aside, even a $50 OEM Android tablet has better UX and usability than Windows 10 without mouse/keyboard.)
It's the complete opposite of how iOS/Android tablets operate.
1) they wouldn't have replaced its Core-based Celerons and Pentiums with Atoms to increase profits - and they sell those for $110-$160
2) they wouldn't have licensed Atom IP to Rockchip and other low-end chip makers, essentially pursuing an ARM-like IP-revenue model (for which they would've made peanuts and didn't pan out anyway).
Intel licenses allot of things they still have ARM licenses and probably even Power PC ones.
The gist of the Digital Semiconductor acquisition was that DEC had kicked Intel's ass in circuit design but turned out to have insufficient volume to be profitable, so Intel and DEC agreed to move DEC onto the upcoming Itanium and DEC could get rid of their bleeding edge but money losing semi business. Alphas and XScales were both remarkable feats - on high performance and low power end, respectively. DEC agreed to kill Alpha and move to Intel's Itanium - but Intel had no existing ARM style product at the time.
I don't think Intel really wanted ARM and so made half-hearted attempts at selling it.
Otellini said Intel passed on the opportunity to supply
Apple because the economics did not make sense at the
time given the forecast product cost and expected
volume. He told The Atlantic, "The thing you have
to remember is that this was before the iPhone was
introduced and no one knew what the iPhone would do...
At the end of the day, there was a chip that they were
interested in that they wanted to pay a certain price
for and not a nickel more and that price was below our
forecasted cost. I couldn't see it.
"It wasn't one of these things you can make up on
volume. And in hindsight, the forecasted cost was wrong
and the volume was 100x what anyone thought."Representing quoted text in italics does not suffer that downside.
(1) was doubtless true. But (2) didn't pan out.
One can find slides from IDFs etc. at the time with Intel trying to buttress its x86-everywhere arguments by showing that Flash ran more consistently on small devices when they were x86-based.
In retrospect, Atom was mostly wish fulfillment. I can't say whether this was an execution issue or whether the strategy was intrinsically doomed to fail.
https://en.wikipedia.org/wiki/Intel_i960
Some variants are still in production mostly for legacy purposes but not the one (MC) I wanted. Too bad.
Now I will have to research about BiiN and Intel iAPX 432. :)
Performance Effects of Architectural Complexity in the Intel 432
https://www.princeton.edu/~rblee/ELE572Papers/Fall04Readings...
The i960/BiiN system improved on that by greatly reducing complexity of the hardware. It was a fast RISC system, had all kinds of error detection, supported HA configuration, and had 432's object-descriptor protections. Object- and page-oriented system. I expect our computer security and reliability situation on Wintel might have turned out differently given what highly-secure systems did with x86's shitty segment/paging/ring combo and HA with lock-step.
Note: A Slashdot article on legacy systems once knocked the F-35 for using old i960 CPU's. Thought that was stupid idea. I think designer was probably thinking too many steps ahead with the market ruining it for him/her.
[1] http://www.cs.washington.edu/homes/levy/capabook/index.html
[2] http://craigchamberlain.com/library/blackhat-2007/Moyer/Extr...
Passive avoidance: Customer doesn't even think to consider what products Intel is offering. Active avoidance: Tries not to spec Intel products when possible.
The internal politics part comes from this scenario. Say someone makes a good proposal, spend $20 million. In return a get product line worth 100 million yr gross and $40 million net. Manager thinks, $100 million a year isn't going to get me a VP position at a $50 billion dollar company.
You could also be describing execs at Apple.
They didn't have any choice. Itaniums were big, and their defect rate on newer processes was too high to get any yield on chips that size.
Microsoft dropped support for IA64 only a few years ago with Windows Server 2008 R2 being the last OS that supported it due to the limited market share. This was quite long after Intel has killed IA64 internally on it's own.
To sell off a division that sold low-power chips for tiny margins in an industry that never, ever, ever explodes (until it did) was just plain normal. He would have gotten pressure and funny looks from the industry and shareholders if he didn't sell it off.
7 years ago they sold their Imageon GPU to Qualcomm, which is now going gangbusters inside every Snapdragon-based phone.
1 http://appleinsider.com/articles/15/01/07/after-intel-spent-...
2 http://www.cnet.com/news/intel-doubles-down-on-mobile-with-1...
3 http://www.digitaltrends.com/computing/intel-to-chinese-elec...
Latest Atoms are almost usable, and you can get passable quad core 4GB ram/64GB flash tablet for $160 (Chuwi Hi10).
More complicated is that Intel came into the mid-90s with several "next generation" architectures, some that weren't solely developed at Intel. XScale came from Digital and had IP from ARM. i960 was a joint venture with Siemens. And they still had the i860 which by this point it was clear was never going to meet expectations in the market. So when the x86 folks said "Intel should put all it's wood behind our arrow", politically they were fighting groups that were weakened by NIH or who'd already for all practical purposes failed. Probably didn't hurt that Pentium was doing quite good and P6 was looking good on the horizon.
So if you go back ten years and say "what if Intel did this" (which in that case was making a processor for Apple that Apple was paying maybe $20 each for, estimating on the very high side), it is over simplified to just imagine that it's additive. Intel has been rolling on profit margins that the hyper-competitive ARM market can only dream about. It may be time for them to adapt (and arguably they have been), but those 12,000 didn't lose their job because Intel didn't do something different ten years ago. They, and thousands others, might never have had an Intel job in the first place if Intel made different choices.
I think the big mistake PC makers are making right now, is that the PCs they make aren't improving from generation to generation for their mass market products. Sure the processors aren't doubling in MHz like they used to, but the rest of the machine isn't improving either. If I go into a shop with $3-400 today and buy a laptop, the machine I get is the same as I would have gotten 3 years ago:
1. 768 line display
2. 5400 rpm hdd
3. 2 GB of ram (4 if I'm lucky)
4. Similar weight
5. Similar poor battery life
6. loads of crapware.
The pc manufacturers aren't pushing hardware manufacturers to improve the cheapest spec. Why don't cheap new laptops have greater DPI on their LCDs than 3 years ago? Because manufacturers haven't changed their main production lines. They are saving money on retooling, but on the other hand their product isn't improving, and now they're paying the price. Apple is doing the same thing with their Air line, which is only improving the processor generation, it has the same body and screen as years ago.
If manufacturers improved their cheapest line every 3 years, people would see enough of an improvement in their price range to buy a new machine every 3 years like they used to.
2014 Intel Core i7-4600U 1718 [0]
2015 Intel Core i7-5600U 1677 [1]
2016 Intel Core i7-6600U 1847 [2]
Two things stood out:
1. Single-thread performance actually decreased in 2015 for some bloody reason.
2. The single-thread performance of the 2016 flagship CPU is only 7.5% higher than the 2014 CPU, a negligible amount when it comes to purchasing decisions. Much of this improvement probably comes from the faster DDR4-2133 MHz RAM (vs DDR3-1600 MHz) rather than the CPU itself.
[0] http://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7-4600U+...
[1] http://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7-5600U+...
[2] https://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7-6600U...
1. 3000x2000 display, touchscreen
2. 512GB SSD
3. 16GB RAM
4. Incredibly light and thin
5. 10 hour battery life
6. No crapware
Of course you'll pay for all that, but it shows there's definitely improvement happening at the high end - other manufacturers are getting there too, though they've got a way to catch up with MS (the HP Envy and the high-end Lenovos were both a lot nicer than anything I saw 3 years ago). Maybe the improvement isn't trickling down as much as it should, but there's room for it to - honestly I think you're forgetting how much more we used to pay for laptops. Time was when $1000 was a basic model. It's just that these days shops push the budget end a lot more, probably because that's what customers want.
It didn't come with Windows?
I'm waiting for cheap usb-c powered laptops before buying any more
Intel, because it charges so much for its chips, making it a big part of a laptop's BOM - like up to 40%. ARM chips are more like 10-15% of a phone's BOM.
OEMs for putting up with it. And consumers for always wanting the fastest (and more expensive) processor in a laptop at any given price, over virtually any other specification.
It's kind of how phones never seem to get more than 2 days of battery life, and bast majority only get 1 day. They value too much chip performance or big screens or high resolutions before they value battery life. There are $200-$250 phones out there with 6,000 mAh batteries and 720p screens, but they come from noname OEMs and most people aren't interested in them either.
You're also a bit disingenuous on your laptop complaints, the best selling laptop on Amazon in that price range has a 1080P screen, 4GB RAM, a decent i3 Broadwell (which is faster and uses less power than its equivalent Sandy/Ivy Bridge processor you would have gotten 3 years ago). Crapware is only a valid complaint if you choose to use Windows and are incapable of taking an hour to install a clean copy when you unbox it.
Isn't this most people?
I did it last week on a laptop that I put in a new SSD in and hasn't had a Windows install in over a year (but came with Windows 8 installed when I bought it). I really don't think Microsoft would go out of their way to make this possible if it was against their terms.
Assuming you get a Laptop that actually works with GNU/Linux.
The only laptop that I still run with GNU/Linux on is an Asus netbook, which was explicitly sold with Linux support.
Guess what, it took more than one year for Ubuntu to properly support its WiFi chipset and I was forced to use a network cable if I wanted any form of networking.
My first GNU/Linux kernel was 1.0.9 with Slackware 2.0, so it is not I am that dumb in GNU/Linux land.
Nowadays I stop bothering and use Windows on all other laptops that I have.
We are nearly at the point where we don't need to give all of our information to someone else just to have it available to us from our cellphones and laptops.
Instead, these companies turned to a standard called ARM. Created by a once-obscure British company, it was designed from the ground up for low-power mobile uses.
Nope, their price budget required plastic instead of ceramic packaging, which had a 1 watt power budget. They were sufficiently conservative that it ended up dissipating 1/10 of a watt. The usefulness for mobile applications came later.
On the other hand, if Intel turned down an offer from Apple to supply the iPhone CPU, well, that sounds like a mistake. Then again, it's such a different business that it's not clear it would have worked for them, especially given the opportunity cost. So different, their FPGA aquisition Alteria is still having their lower end more price sensitive chips fabricated by TSMC, apparently because Intel is just too expensive for that market.
And Apple could well have changed to ARM later, Macs are now on their third CPU architecture.
True, but not quite the whole story.
Some enterprising engineers wrote a cellular protocol stack in ARM assembly language that let everybody use a far cheaper core than anything else.
Very quickly, ARM became entrenched in the feature phones. Then, the evolution occurred to smartphones, and ARM was already in the phone.
But the chips in the PC had the best margin by far. So the more of those they made, the more profitable they became, and when the chip recession was in full swing in the late 80's and early 90's that is what they kept, shedding all the rest.
In the early 2000's when Moore's law ran right smack into the power wall, Intel was betting they could have an "enterprise" line (Itanium) and a "desktop" line (Pentium), and an embedded line (8051). They guessed wrong and for a brief time AMD's Opteron was kicking their butt. But once they realized the writing on the wall they realigned around 64 bit in the Pentium line and got back on track.
The problem with the ARM assault is that unlike AMD, which could be killed by messing with other users of the chipset and patent attacks and contract shennanigans, killing off someone making an ARM chip does nothing but make the other ARM chip vendors stronger. And they can't kill all of them at once. And worse, to compete with them they have to sacrifice margin on their x86 line and that is something they have never done, it breaks their business model.
Its a real conundrum for them, they don't have a low power, reasonable performance SOC architecture to compete with these guys. And that is driving volumes these days. Further, the A53 (ARM 64 bit) killed off the chance of trying to use 32 bit only ATOM microarchitecture chips in that niche without impacting the value of the higher end Pentiums.
One of the things Web2.0 taught us was that it doesn't matter how "big" the implementation of a node is if your going to put 50,000 of them in a data center to run your "cloud." Ethernet as an interconnect is fast enough for a lot of things.
It definitely makes for an interesting future.
Someone who was in IT before MS and Intel showed up said MS and Intel got big even though they did not have the best technologies. They got big essentially with smart alliances and good business practices (which included messing with competitors, some practices that can be called dirty).
Microsoft and Intel got big for one and only one reason: IBM chose them as suppliers for the IBM PC. Had IBM used their own in-house chip or licensed CP/M as operating system computer history would have been different (the Kildall link is especially interesting, Microsoft got really lucky here):
https://en.wikipedia.org/wiki/IBM_801 https://en.wikipedia.org/wiki/Gary_Kildall
For much of the 80s the Intel chips were inferior to other designs like the Motorola 68K. That is why the Macintosh, Atari (today only known for Pac-Man, but yes, Atari made computers rivaling Apple) and Commodore Amiga used more powerful Motorolas.
http://www.skepticfiles.org/cowtext/comput~1/486vs040.htm
But the "IBM-compatible" architecture won despite its inferiority through path dependency and the clones driving price down.
I do agree that IBM's choice of the 8088 for their original PC was one of Intel's greatest design wins, but at the same time it was the execution of both Microsoft and Intel in their focus on pursuing the business market which made it truly successful. The "hobbyist" microcomputer market in the late 70's and early 80's was scattered. With small wins all over the map.
And therein lies Intel's quest to become relevant again. Make more bloated languages so more processors can be sold to servers :/
And it was better than C in terms of safety.
What influence did Intel have on the design, besides paying for it?
A company with a profitable niche and a profitable technology will wind up with high internal costs. That's fine in their main business because they have a profit margin to play with. But it is surprisingly hard to trim back that "fat" to go after much lower margin revenue with a cheaper technology. (Fat is in quotes because it isn't really fat. It is necessary for the high margin business.) It is common to try, and to conclude that it is a failure.
That is why Intel made this mistake.
ARM has volume, but x86 has profit.
Intel will continue with the high profit margin right until the chips don't sell. And then Intel gets to go out of business.
Or, it can switch to selling ARMs and be a fraction of the company it is right now.
Intel's revenue last year was a record $60 Billion. If Intel could get $30 per ARM chip and didn't have to pay a royalty to ARM (ha ha) and had every single iPhone sold last year, that would only make them a $6 billion company. In reality, Apple would never pay that and ARM takes their cut. So, at best, Intel would only be about a $1 billion revenue company, and, that was only last year when Apple moved 200 million iPhones, the previous year was only 100 million.
So, Intel would be roughly 1/10th to 1/100th the size it currently is if it manufactured ARM's instead of x86. Yeah, I'm sure managers inside Intel are lining up for that business decision.
It is going to be better for Intel to glide to "irrelevance" for a very long time rather than switch to making ARM's.
I dont think that's actualy what anyone is suggesting.
So, you're telling me that a fab line manager is going to reduce his profit margin by 10% just so he has a fallback when the x86 market collapse and decimates Intel?
This is like finance guys before the stock market crash: "I can be contrarian but it does me no good. If I'm right, my company is so invested in the stock market going up that my company is dead anyway. If I'm wrong, well, I look like an idiot and don't make money. So, I'll just try to make money and cash out."
And how exactly would them (Intel in this case) selling commoditized, low margin, products make it better for them?
Isn't it even worse, even quicker, for low margin players when their stuff doesn't sell?
He's gone, and Apple has stopped doing that. Apple is now losing marketshare. (Android is estimated at 82%.) Their app store is globally under half of the market. They are projecting a year over year decline this quarter.
It probably won't be visible to the untrained eye in the next 5 years. It won't be missable by anyone in the next 10. But Apple's best days are behind it.
Actually they did just that with the Apple Watch -- which added ~6 billions to their revenues, and even as an early v1.0 eclipsed all "wearables" to date. http://www.cnet.com/news/thanks-to-apple-watch-smartwatch-sa... http://www.macrumors.com/2016/01/26/apple-watch-apple-tv-rec...
And their services dept doesn't do that bad either: http://appleinsider.com/articles/16/04/20/as-a-standalone-co...
They've also keep improving the Apple TV (people who haven't followed Apple forget how slow and incremental the rise of the iPod was -- from 2002 to 2007 people gathered at Keynotes to cheer if it got silly features like Wifi or a color screen or some smaller sibling), working on a car, and other things besides.
>Apple is now losing marketshare.
Barely -- from April 2015 to now, they've gone up and down, some quarters winning over Android, others losing. http://www.comscore.com/Insights/Market-Rankings/comScore-Re...
Besides I've never understood this "let's pin Apple, a single company, against the whole of the industry put together". They've never had the "most" market share -- just the most of the most lucrative (higher end, high margins) segment of the market.
>It probably won't be visible to the untrained eye in the next 5 years. It won't be missable by anyone in the next 10. But Apple's best days are behind it.
I think I've read that again, in 1997, 2000, 2002, 2004, 2006, 2007, 2009, 2012, 2015 etc. A.k.a "Apple is doomed".
The iPod never meant you don't need a Mac.
The iPad never meant you don't need an iPhone (and hardly ever meant you don't need a Mac/PC).
The idea that a device should replace previous devices was never much of a concern. The only thing that qualifies 100% in that story is that the iPhone was by nature also a portable music player -- and if you had one obviously you didn't need the iPod. Apart from that, all were individual lines, with their own strengths and limitations -- not supposed to replace one another.
http://www.bloomberg.com/news/articles/2015-10-05/did-clay-c...
Really it's hard to find a platform that isn't from Intel from the last 20 or so years that doesn't show an influence from them.
The AMD64 usage implies non-compatibility with EM64T..
Colloquially, "x64" works pretty well.
My memory may be a bit hazy, but wasn't one of the demonstrations of the original ARM a program in interpreted BASIC that did in twenty seconds something for which a compiled C program on a 80386 needed thirty seconds? Or something like that?
The first was them trying to understand why the ARM chip they had on a test board was working, despite the fact that the power supply wasn't plugged in. It transpired that there was leakage currents from the keyboard which ended up giving the CPU enough power. (Discussed in a Computerphile YouTube video on the origins of ARM.)
The other was a demo of showing an ARM chip being powered off a thermocouple of the waste heat from a 286 (or maybe 386 ... either way an old Intel CPU)
What made BASIC blisteringly quick on it was that the BASIC interpreter that Acorn had originally written for their ARM-based micros was small enough to fit into the instruction cache on the StrongARM. (According to Wikipedia, StrongARM was 16kb I + 16kb D cache, a Harvard Architecture; whereas the Acorn designed cpus only had a 4kb unified cache at the time.) BASIC programs (which were byte-coded) were enormously faster on the StrongARM.
Not bothering to read the rest. This is entirely 100% wrong. The PC era has not "ended". It's just that we only upgrade every few years instead of every year. And grandma now reads her email on a tablet instead but that was never what PCs were really for.
PCs are still just as much used as ever. We just use other things too, and don't buy a new one every year.
If they can't get this basic fact right then I have no hope for the rest of the article.
Eras are ill-defined, but if you can assert that we once were in a PC era, you must also accept a definition that allows, in principle, for an end to an era.
The dominant platform of its era is the one with the greatest user and developer person hours, sales volume, zeitgeist and so on.
We are in the mobile era.
PCs are now a mature technology, just like cars, and have the resulting long lifecycle.
Smartphones are also getting into this stage.
I guess there are 12,000 other workers somewhere else in the world that now have a job because they get to create what Intel doesn't. BTW, according to past statistics, most of the workers that are now "paying the price", wasn't even Intel's employees 10 years ago.
In the 2000s smartphones and mobile devices outnumber PCs. Intel missed that and so ARM dominates the business.
Maybe that same pattern will repeat again with the rising of IoT and wearables, where smaller and cheaper chips become ubiquitous.
The development of Edison and Curie processors might indicate that Intel is betting on this, gearing up for the next "disruptive innovation".
Or possibly more likely - the next transition which will be much messier, with no single winner.
The mainframe -> mini -> desktop -> laptop -> mobile -> SBC path was about miniaturisation. The devices all provide general computing facilities, but they get smaller and faster and use less power, while the UI becomes more accessible to non-expert users.
I'm not seeing how that translates into wearables, because IoT and wearables aren't general purpose computing devices. They're more like embedded hardware and/or thin network clients.
So I don't think that's where the disruption will happen. It's maybe more likely the disruption will happen in consumer AI, where the UI gets simpler still through speech recognition and NLP, and the screen/mouse/keyboard start to disappear.
My guess is traditional processors will become front-ends and glue for hardware AI systems, and the companies to bet on are the ones producing the subsystems. The hardware will follow the same path as general purpose processors, but they'll move from "mainframe" to "SBC" much more quickly.
[1] http://www.sandiegouniontribune.com/news/2015/sep/17/Qualcom...
This has become popular myth, but I don't think it's true. Sure, Microsoft wasn't far ahead of the curve on the internet, or they wouldn't have been developing proprietary online services designed to compete with AOL and CompuServe.
But the Bill Gates "Tidal Wave" internet memo was sent in May of 1995! Netscape's IPO was still months in the future, Amazon had maybe a few million $ worth of revenue, and essentially every other web-based company still in business today hadn't been founded yet.
IMHO, billg's memo should be seen as an example of great prescience, rather than a belated corrective maneuver.
Love it or hate it, Intel still has the right technology and products to appeal to a broad market and make good money. One cannot expect to win every market, you can try if it makes sense and you should know when to walk away.
"Monopolist Missteps and Loses Monopoly Position"
http://appleinsider.com/articles/15/01/19/how-intel-lost-the...
This ate up valuable chip real estate, RAM bandwidth, thermal overhead etc. Worse it cemented the idea that Intel was crap at graphics while slowing down the PC upgrade cycle.
Note that ~15% of Steam Players are gaming on Intel iGPUs, as awful as they are.
Intel had the ability to put more transistors on the same die size with the same power requirements. This was long after they reached thermal/clockspeed limits (with the P4). They started putting additional cores in there and bumped up the L2 and L3 caches, but there was still space left on the die.
What do you do with those extra transistors? It would be absurd to "leave them blank" as you are basically throwing money down the drain.
The iGPU needs to be manufactured anyway. And with memory controllers integrated into the CPU, it seems rather complicated to push the iGPU off the chip.
> and more RAM bandwidth
Cheap iGPUs on Motherboards would recycle the CPU's memory controller anyway. Did you work with computers in 2006 or so? Its definitely cheaper and more efficient to just integrate everything into the die in contrast to the designs of the past.
Er… you mean like every single mobile SoC?
Which means if one part is defective you can use a different part boosting yields. But, they also get much lower latency.
Now recent Phone Chips have included on die GPU's but phones are so power limited it's not an issue.
HBM is only being used on AMD's R9 Fury Graphics Card, as far as I'm aware. A rather high-end solution that simply doesn't exist in the phone market right now.
PS: I could get more accurate numbers, but with such a wide range of latency number that's pointless.
As I noted before: Mobile CPUs are using standard DDR3 or DDR4 RAM, on a motherboard drawn with copper wires to the CPU. The ONLY use of "3d stacked memory" that I know of is the HBM RAM of the AMD Fury X.
http://www.legitreviews.com/wp-content/uploads/2015/05/HBM-D...
I'm laying counter to your fact-claims. Only high-power GPUs have stacked RAM right now.
Anyone who has opened up a phone can see the DDR3 or DDR4 RAM on the phone's motherboard, with the phone SOC designed to go "inefficiently all the way" to the DDR3 RAM over "long latency" copper wires.
Read the spec sheet yourself if you don't believe me.
http://system-on-a-chip.specout.com/l/1107/Qualcomm-Snapdrag...
Anyway, I am referring to packaging like: http://cdn.arstechnica.net/wp-content/uploads/2013/04/ASIC_+... Yes, HBM also does more or less the same thing. The issues are twofold heat which is vastly less of a problem in the mobile world so with HBM it's a stack of ram chips next to a processer, with Mobil the CPU is under what might only be one RAM chip and they call it SoC.
http://arstechnica.com/gadgets/2013/04/the-pc-inside-your-ph...
Note: CPU's are already 3d with many layers depending on how you count.
Anyway after die size, heat, latency, and manufacturing costs are really the largest factors. Some PC ram chips run hot enough they really can't be stacked. You can slow them down and trade latency for total bandwidth. ed:Less of an issue with active cooling and a heat sink. However, they can call this HBM, because you can have more chips so even if each chip is slower overall you get higher speeds.
PS: Don't forget HBM is designed for a GPU/CPGPU without a large cache. CPU's have a huge cash which let's them have very different approach aka lot's of cheap but slow RAM. Benchmarks show faster ream really does not add much for a modern CPU with most workloads. You could use this for a CPU but it's harder to swap in more ram.
I am pretty sure that if keeping the memory controller, northbridge, and iGPU separate, Intel would have kept these things separate.
My bet? I think the cheaper manufacturing costs associated with motherboard construction beat-out the slightly increased cost of tighter integration. But I'm not an Intel Engineer. Just a hunch... considering that Apple A9, Qualcomm Snapdragon, AMD A10 "APUs" , and Intel have all decided to integrate everything together.
> better thermal envelope processing speeds
What, are people actually using iGPUs for difficult tasks all the time? As far as I can tell, most people surf Facebook all day and ping the CPU.
iGPU isn't slowing down anything with Facebook surfing. Besides, a crappy integrated GPU on the MoBo would still use up RAM bandwidth.
If anything, integrating the GPU has made the process cheaper for Intel. The Northbridge, Memory Controller, and iGPU are all integrated into modern CPUs, because Intel has done the number crunching and has determined that its cheaper to do it this way.
It's much easer to cut rectangular chips though. Still, even if the used the same overall design and just ignore any defects over X area would boost yield.
Thus main memory is much slower than GPU bandwidth. So, now your on chip GPU is memory starved and reducing the bandwidth avalabel to the CPU. There are some advantages if you want to pass stuff back and forth, but the standard rendering pipeline is designed for GPU's.