Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
Ok, let's see: Intel already owning a 16% share of the baseband market which is about 50% of Qualcomms market share. Now, Intel about to overtake Qualcomm in 4G technology but the winner in the end will be Qualcomm and Mediatek. Sure, whatever thesis suits your investment ...
The actual point is integration. The baseband needs to be integrated into the SoC for cost reasons. Intel is about to do that, a bit late, but on their way. They don't support all worldwide bands like Qualcomm does, that's true but about to be fixed. That is the reason why they bought the Fujitsu RF team. I guess they'll have all that by next year, including integrated GPS.
On the other hand, ever realized how the mobile market isn't driven by the modem makers anymore? The investments involved in new standards are very high and the carriers seem more and more hesitant to make these investments. The whole 4G development slows down and becomes a commodity where Qualcomm has nothing to offer compared to the competition. I see a price war coming there where the external modem guys will be finally killed by those who integrate the whole thing into one packaged SoC. That's exactly where Intel will shine, with the leading edge process and the lowest cost position since no royalties to pay for the processors, eventually graphics and their leading edge process which is still following Moore's law.
Yeah, and you are among those 10 guys that are actually willing to pay the same price for getting less - no legacy compatibility and slower application performance (yes, Silvermont is faster at SpecInt). Dream on ...
TSMC is not mobile, nor PC nor any specific product. They are a foundry as you may know and they prduce a large variety of semconductors for their customers. They are, for instance, producing PC chips for AMD, graphics cards and a bunch of analog and mixed signal chips.
TSMC is a good company and it is not as if Intel is directly competing with their business model. I can still remember as noone was taking them seriously since they "only" produced for others and had no own designs.
It will have a big impact on TSMC in case Intel gains significant(high end digital) design wins like the one from Altera. Actually I did own TSMC shares until spring last year but I sold them once the Altera deal with Intel has been anounced. In addition, Samsung and Globalfoundries follow them closely, taking more and more share from their 28 nm (sweet spot) business.
It's not as if Intel will produce everything TSMC produces today. Intel is specialized on high performance (and nower days also low power and high density) digital designs. Intel has a technology lead there but when it comes to broad support for all variations of semiconductors and the associated process technology needed, Intel can't (and shouldn't try to) compete with TSMC. That's a completely different story.
That doesn't mean TSMC is unaffected by Intel. In case Intel gains significant share in mobile and the steal some of the high end digital customers like Altera, TSMC is going to suffer from that just like Qualcomm and ARM will.
The issue is that there is an oversupply of semiconductor in this world since every shrink increases the capacity of the fabs. This is also the reason, despite the high costs for new fabs, why there are only a few foundries and Intel left with leading edge processes. They can already produce a big chunk of all the semiconductors needed in the world.
A final remark regarding this american company stuff: Home bias is usually considered a bad thing for your investments.
Ok, that may happen. Not a lawyer either but it certainly is not allowed to distribute devices with patent violating parts. Still it would not be Apple who is sued but Intel, just like Broadcom sued Qualcomm and not the device vendors. It would affect Apple in that scenario if Intel wouldn't find an agreement soon enough, you got a point there.
Please explain: How can they sue Apple when their patent is violated by Intel's IP? Would they sue Apple when the used Qualcomm modem violated their patent? It's actually different if Apple would implement its own modem instead of integrating it as IP. That is a counterargument against Apple developing its own baseband.
I was saying Intel would give their modem to Apple as IP. That would mean for mobile technology, Samsung would have to sue Intel and not Apple. Intel would also have to sue ARM for breaking any patents in one of their standard cores, e.g. in a Mediatek. That is very different to what Qualcomm does, since they have their custom processors/IP.
Oh, ok, I am not getting it! What makes you think Intel has no significant patents in mobile? It's not that Infineon was new to that business. Actually, they had quite a market share for a while and many new technologies are based on those methods and patents. Secondly, Intel has patents in other mobile technologies, e.g. Wifi, which are used in almost all mobile SoCs.
Taking your specific example, Samsung: Remember, Samsung has a foundry business? How do you think could Intel screw them with their process patents? Intel is not exactly the company to mess with when it comes to tech patents. But probably Mediatek will sue them badly ...
And you think Intel has no patents to hold against? CPU technology probably? Qualcomm Krait is a custom CPU core. You really think Qualcomm would risk that?
Yeah, Mediatek will help them out of their misery. They still seem not to get it. What is the value proposition of Windows RT compared to Android? Less Apps? Higher memory consumption (at least Flash)? They can give it away for free (actually they are) but it still offers no reason to switch from Android. In addition, they are not even making money with it.
Microsoft should instead think about support for the most important windows applications to easily port them to x86 Windows for tablets. The price won't be much higher and people get applications they cannot find on Android or iOS. I really hope Microsoft will lose market share in the PC space also, for the mess they caused there for decades. They are among the worst software companies in the world with buggy, uninspired and slow software. I am really hoping for Linux to free the desktop as well - this would also be for Intel's benefit.
That's what I thought initially, when i heard of this rumour. What about Intel fabbing Apple's upcoming chips at 10nm with Intel's modem as integrated IP? That makes a perfect business package for both sides. Intel nicely fills a big chunk of its fabbing capacity (improving its own margins at the same time), sells its own modem as IP while helping to integrate it (and maybe protect their IP from external insight since it doesn't need to leave their fabs).
Apple on the other hand would gain the most advanced process for a performance/density/power lead towards their "competitors", justifying for a nice premium for their hardware. Maybe then it's not about having 64 bit but having 10nm chips in your phone? Apple is known to want the best - Intel can deliver that.
Makes me bullish but sadly we have to wait and see.
There is one slide on page 23 that may be one of the issues Intel has today with further integration of connectivity: 3.3 Volt. Probably they needed dedicated transistors they haven't had the process steps needed for. Just a guess and this alone maybe not the reason why it takes them years to get that but probably part of the problem (wild guess, nothing else).
Anyway, they'll have to pay 8 cores instead of the previous 4 or two and they have to pay a premium for 64 bit. I think it adds significantly to the bill.
Ups, looks like I am going to lose my ARM shorts - again ... That's a lot of royalties for ARM to make. It is said that they are making 60 cents per high end core with 64 bit instruction set (probably half of that for an A53). That would be around 3$ per chip - good for ARM and good for Intel (higher prices to compete with), bad for Qualcomm. I think Mediatek and the others will have a very similar design as well - not sure if Qualcomm will be that successful with it. ARM on the other hand, will only lose income when Intel grabs a significant share of the market, making up for the increased royalties ARM is receiving in the high end (big.Little and 8 core designs).
From the article, I didn't really get what the 810 looks like. Is it a big.Little A53/A57 quadcore, meaning four A57 and four A53 on one chip? That looks kind of desperate. big.Little is crap in real life since switching between the cores is costful and it seems that A57 is, again, just like A15, a power hog. Since they'll probably be able to run all 8 cores simultaneously it will do very good in benchmarks - for a very short time until you burn your fingers
Anyway, really looks like Intel was very lucky to have 64 bit cores, even though they only constructed them because Silvermont was also needed for operating Windows. But hey, since there are people who think they need 8 cores in a mobile chip I guess there are people who think they need 64 bit on them.
The big question now is: Will Intel's Broxton be so much better that it overcomes the stigma of not being yet another pseudo 8 core (big.Little) chip? I really hope so and I bet real life performance will be way beyond that of the competition - but they need to sell that as well. Silvermont is already in the ballpark of A15 and A57 is not much faster than that. Should be possible, but benchmarks will be worse I guess, since beating an 8 core is hard to do for a quadcore.
Good for me. I am also long Micron (and already gained about 70% with that). Intel and Micron are kind of a dream team - last men standing.
Easier doesn't mean easy. Actually it's the difference between "almost impossible" and "hard, but can be done". Just have a look at Windows RT. Why do you think it is not possible to run all that legacy apps on it (despite ARM cores being a bit slow for that)? Well, many/most Windows applications are also precompiled, just like it is done for iOS (I'm not going into the details of .net right now). Even if ARM cores would have an advantage towards x86 (which they don't), it would still not happen that applications would be ported from one day to another and emulating the legacy apps would be a huge pain on Windows.
Android is, as I said, very different, since applications are not precompiled but compiled just in time. This enables the OS to use the compiler that is needed for the given plattform it is running on.
So, what is so hard about that: The whole OS needs to be ported (well, based on Linux which originates from x86, that is not too hard to do). But, in addition, you need your own drivers and, probably the trickiest part, they needed to emulate native ARM code, which is especially crucial for optimized code in games etc. That probably was the hardest part and they did well as I can confirm from using an Intel Android phone.
If Android wouldn't be Java/Virtual Machine based, Intel probably would be out of the mobile business once and for all - so, that actually was really close ...
Since you're still here (wanted to ask that question earlier), could you provide some links? I am always looking for more information and it would be a great value you could add to the discussion!
Sorry, but to my knowledge, this is wrong. As far as I know, all ARM application processors (i.e. A7, A8, A9, A12, A15, A17) are binary compatible, meaning they can execute compiled binaries on any of these. The same is valid for Apple's A processors, which are based upon the same common ARM instruction set (I think it is ARMv7). Also for the 64 bit cores, the instruction set is downward compatible (ARMv8) meaning that they can execute ARMv7 binaries. It doesn't work vice versa, meaning that you can't execute 64 bit compiled applications on an 32 bit ARM core, which makes kind of sense.
The real difference is, that Apple doesn't use a virtual machine like Java does but instead uses precompiled applications, like in the old days. That is good for performance but it's bad when you want to change the processor architecture.
Android on the other hand is based on a virtual machine (Java) which is fed with so called bytecode or intermediate language which is compiled just in time (at time of execution) to the native instruction set (ARMvX or x86/x64). This cost some performance but it makes the program more or less independent of the processor architecture, which is big luck for Intel.
There is still the possibility, also on Android, to write native code. This is often done for very specific/low level or optimized code. In most cases, these are games. Intel uses emulation for those parts and provides the possibility for newer apps to precompile those parts for the ARM and x86 instruction set and deliver both with their apps. That was the hardest part to do on Android and Intel did it very well. Almost all apps are running on x86 and even games are mostly running well. Good work by the Android programmers at Intel!
Um, it's an x86 CPU. How can that be compatible with the ARM instruction set? Technically, this should be possible (would bloat the core though) but I doubt that would be legal. Remember how Intel only gave two x86 licenses to others? One is hold by Via I think and the other at AMD. I guess this is similar with the ARM instruction set. It's some kind of IP.
Porting the OS is not the issue. The existing apps need to be ported as well and that is something Apple can't do since they don't own the sourcecode. The app developers have to do that and this won't happen from one day to another but takes years instead until everyone released a ported version. In the meantime, those apps have to run on an emulator which is performance crushing. This may make sense when Intel provides CPUs with twice or more of the performance (at the same power envelope) as Apple can do. Something currently out of sight I'd say.
Thanks for the link. It's only scratching the surface but that doesn't matter here. We'll probably never know about the details of that - not meant for the public in any case.
In general, Intel may be (well, actually is) ahead of TSMC regarding digital circuits but I think they are way behind regarding analog stuff. TSMC needs to serve everyone, from digital to mixed signal to very special analog chips. That's one of the reasons I have some doubts about Intel's foundry perspective. Sure, there is a lot pure digital chips they could make and they should take that business if possible but I doubt it will be sufficient to finance their upcoming fabs. One shouldn't undererstimate TSMC - they are not new to this business and they are doing stuff for many years Intel probably never did before.
According to that illustration, there's "only" ADCs and DACs needed as analog circuits in the baseband controller. This may sound easy but for very low noise and low power consumption, those can be quite tricky. In addition, this figure may not be accurate up to all detail. It shouldn't be impossible for Intel to do, though. I certainly have no clue what it takes to build good ADCs and DACs in 14nm. By the way, there may be some surprises for those fabbing at 16nm TSMC as that won't be a piece of cake for them either, I suspect.
Besides you can see that a lot of know how for those new RF standards is actually packed into the RF frontend chip which is also the reason why Intel needed to buy those Fujitsu guys. I guess that was a good buy (other than that TV and IoT BS).
Yep, that's what I am saying. Nice to have someone who knows how to do analog circuits here on this board. It's a rare species these days but they are often key to the really difficult stuff. I only do digital design but some of my colleagues are analog designer (experienced guys, I like them a lot).
Anyway, regarding analog, bleeding edge processes are not the best. Analog circuits don't shrink really and the latest process often miss options that are needed for some analog designs, such as BICMOS, high voltage transistors etc. I guess the best analog processes with the most options these days may be 130nm or 90nm. 14nm is not what an analog designer dreams of at night (as nightmare maybe). Mature processes are often best for such designs.
Anyway, as long as the digital designs are synthesizable, which I guess most are these days, especially ARM based ones, it's not an issue to port them. It's analog designs, hand layed digital designs (like Haswell maybe, for speed reasons) and memories that require significant work to be ported.
I really don't get your point. Infineon used ARM cores for their cellular modems but that is not the problem porting to Intel's own processes. That's an automated process, at least for the low clock rates needed for modems. It's the other stuff in the baseband controller that may be the problem, especially if it involves analog circuitry of any kind. I really don't know what a typical baseband controller is comprised of in detail so I can't tell you whether this might be the reason. I can imagine though that porting existing analog circuits from a planar process to finfets is not a straightforward thing to do.
@Ashraf: Maybe Intel tells you that this was just a question of timing and so on but I reall doubt that this is the truth. Sorry, but I guess Intel won't tell an outsider like you what the actual reason is. If it was just timing and planning, why are they only planning to integrate the modem until the end of 2015 if it was a matter of using an existing design and simply synthesizing it for their own process? There's something we don't know and it seems like a major issue to me. If porting their SoC to TSMC is easier to do than porting their modem to Intel then it is definitely not about planning.
We were talking about modems and their integration, not about some ARM design. If it's a mere digital design that is synthesized, I don't see any issue transferring that from one fab to another, including Intel's. That's a non issue.
Things are very different for digtal designs layed out by hand (as far as I know not or rarely done in the ARM world) and especially for analog circuits. Such designs take into account the various process characteristics and those change from fab to fab and from node to node. Porting analog circuits from one node to another is no automatic process but requires resimulation and often additional tuning and changes in the circuits up to the layout.
Now the real question, at least I don't have an answer for: Are integrated baseband controllers pure digital circuits that are synthesized or are there analog or hand layed out parts that require adoption? Is there anyone on this board who can answer that?
Just to explain: You certainly need analog circuits for an RF system but those are never, also not for Qualcomm, integrated into the SoC. It would be hell on earth to integrate such a sensitive circuit together with those horrible EMC distorters like CPUs and GPUs.
That's the second choice they have. Sure they can fab and sure it helps them pay the new fabs, but the margins are simply lower. If Intel gains significant share in mobile (talking about maybe 20-30% of tablets and smartphones) they'll get TSMC's share, Qualcomm's and ARM's (+ graphics and other stuff maybe). That's a lot more than just TSMC's share and that is also the reason why Intel doesn't have issues competing in price. Just consider how Qualcomm can charge its ~50% margin compared to Intel's 60% (and TSMC has a margin in addition at around 30% or 40% if I remember correctly). Mobile chips aren't as cheap as they seem.
The best is when Intel does both. Win significant share in mobile and do some foundry business for leading edge customers that are willing to pay a premium, like Altera currently is. Apple is a perfect match here, not Qualcomm. Apple doesn't currently have an integrated modem (where Intel is struggling with right now for whatever reasons) and I am pretty sure that iOS will never be x86 compatible since porting that is much harder than doing this for Android due to its natively compiled programs (compared to Java virtual machine based Android). Therefore Intel would need to provide a multiple of the performance/watt than Apple can with ARM - something not in sight, before Apple would consider to switch from ARM to Intel, like they did from Power PC to x86 for MAC many years ago - the emulation layer needed for the transition is performance killing. On the other hand, Apple could well stay at the performance top when using Intel's bleeding edge processes and they'll keep the Android guys on distance to charge their premium for their devices. Intel on the other hand has no issue since Apple is not competing with their Android chips. The cooperations between Intel and Apple could be extended in a way that Intel provides its modem technology to integrate into Apple's chips (further harming Qualcomm) and on the other hand, Intel could protect the MAC business via contracts. That would be a win-win. Intel fabbing for Qualcomm is not nearly as interesting and realistic until Intel gives up on mobile completely.
I hope the rumours about Intel fabbing for Apple in 10nm are true. That, in addition with gaining significant share in mobile and maybe a more healthy PC market (e.g. due to Chrome OS replacing crappy Windows) is my super bull scenario for Intel.
You are absolutely correct but that is not a disadvantage for Intel, it is actually an advantage. Fix costs are very high for high end fabs (maybe around 5 billion each). Intel finances this with their PC and server business. This won't be possible anymore when fabs become more and more expensive with every new node. That's what they need volume for and mobile fits best, with the best margins. The thing is: Intel already produces chips with great margins in their own fabs which help, even when fab utilization is low (Intel denies that but I consider it a lie), to finanze those fabs. All the additional volume from mobile, even at low margin, actually helps to increase the margins of their high end products. This is the great thing about Intel: Even if they flood the market with low margin products in mobile, it helps them to pay for their fabs and margins for their PC and server processors get better at the same time. In addition, once they are ready, they can have nice margins in the high end in mobile in addition, which currently Qualcomm is able to charge as well. That's a very nice perspective but it is critical how they execute in mobile - something they know and they are starting to be very agressive with.
Yes and no. First of all, Intel also has depreciated fabs at previous nodes. They used them for Medfield and Clovertrail for example. The thing is: Intel clearly stated that, compared to TSMC and others, their cost per transistor curve follows Moore's law just as it did in the past, even for 14nm and beyond. That's important because that makes 14nm actually the least expensive option for Intel in mobile - but they need the volume to pay the fabs (fix costs). By the way, also TSMC and the others have to pay for their new fabs and get the money from somewhere. There are also technical reasons why you need smaller processes (and please, no "good enough" argument this time).
Doubt it. They already announced the transition of Sofia to 14nm. The reason for selecting TSMC for the first incarnation of Sofia is Intels modem, which is already made by TSMC. Integrated modem on the other hand is very important for the low cost market. I suppose they have issues with some analog parts of the modem and they need additional process steps for their 14nm process. You shouldn't forget that Intel is mainly a digital company and has not much knowledge in analog or mixed signal designs. They can do it but it's not as easy to do as it may sound. Initially I thought the baseband is a pure digital part (since an external RF frontend is always needed, also for the competition), but that may not be completely true. Maybe there is someone here in this forum with a deeper knowledge of that stuff.
It does make sense in case Intel's CPUs are power optimized for their process and this application. Royalties of maybe 30-60 (they often use multiple processors) Cents for a chip that costs less than 10$ is important for the margin. Concerning delays, I think that is not an issue. Such a transition can be developed independent of the product which it goes into. The firmware is mostly C I guess and porting is not that much effort actually.
When it comes to costs, integrated solutions are the best/cheapest. That's why Intel needs their own bleeding edge modems, so they can integrate them with their own processors and provide an attractive package. A 14nm SoC with integrated comms definitely will be cheaper than a 20nm application processor together with a 28 nm modem (especially when not fabricated in house).
Intel is doing that step with the Sofia successor, at least according to their plans (which are somewhat hot air these days ...). The high end chips with faster processor and modem (probably the Broxton successor with integrated comms) won't be much more expensive (slightly larger die size) and deliver a nice margin if it significantly beats the competition (a big if so far).
In addition, Intels 14 nm process may help to realize a highly power saving modem which is actually more critical for battery lifetime than the most-of-the-time-sleeping processor.
By the way: Someone told me that Intel is about to use their own processors for the modem firmware as well, helping to reduce the costs and power consumption further, since they don't need ARM cores for that anymore. Intel has a great cost potential nobody else in the market can beat. I think that is what many people actually misunderstand since they think Intel can't beat the cheap ARM SoCs in price, which is completely wrong in my opinion (due to less royalties paid, denser process and their own fabs). All they need is volume for that to take off.
Thanks for the infos, Ashraf! Great to have you back!
I'm pretty baffled about Intels execution speed regarding Sofia and Bay Trail entry. Looks like the titan finally woke up - duck and cover everyone in range. I placed a small short on ARM today, hope that will work out. ARM will be affected the most by these cheap Intel 64 SoCs. Intel is obviously pricing it very attractive and there is nothing in the market close to it regarding performance/watt. Who said Intel is too expensive for mobile?
You're mixing that up with Cherry Trail, which should be available by end of this year. There is no 14nm Intel product in the market right now, didn't you know?
Well, first of all the share price is stable since the announcement - I don't see much negativity due to that from the investment community. Secondly, Bay Trail Entry is a redesign. The announcement was made in december, so about 4 months ago. I doubt Intel can make a redesign with such major changes (memory interface etc.) that quickly. Also, before starting the work, management must take decisions, discuss details etc. That also takes time. Therefore I believe the decision for Bay Trail Entry has been taken before the investor meeting where they announced the contra revenues. I think they originally wanted to have devices with Bay Trail first and then switch to Bay Trail entry. Something went wrong there.
By the way: Where are our Merrifield phones? I hope we don't have to wait for them another 5 months ...
Besides: If Bay Trail Entry was so easy and quick to make I really don't get why Intel released Bay Trail first which has the BOM issue. What type of planning was that?
Well, I think most consumers, including myself, wouldn't want that crap even for free. Microsoft is just making bad software for so many years now and people are fed up with it. Why switch from Android with its larger ecosystem and liberate yourself from Microsoft frustration when it's just "also free"?
I'm really hoping for Chromebook liberating the desktop as well - and no, this is no bad thing for Intel in any way. Actually this would increase desktop and notebook sales significantly.
This kind of explains the BayTrail+ rumour. Honestly, I don't believe this is the actual reason why Intel didn't release BayTrail for Android devices. How much sense does it make to announce huge contra revenues and then not to release the product and wait instead to make it cheaper? BayTrail is already late to the party and they should have released it as early as possible. I think they screwed something up, maybe graphics drivers, but I'm not sure.
Come on, wthdik, this really is nonsense, sorry. 28 nm is a mature node for quite a long time already and there are no capacity issues anymore since Samsung and Globalfoundries provide the same nodes. There is nothing for TSMC to delay. Intel will provide the layout and TSMC will generate the masks and start its production. That is a matter of a few weeks. TSMC can't and won't delay anything here.
That's a bad reason to delay a product. They could have introduced 32 bit Android versions first and guarantee customers/OEMs that a 64 bit upgrade will be available to everyone. No reason to delay.
Highly doubt it that Intel would sign such a deal since they are focusing on Android. Also Surface RT 2 is proving that wrong - that would have had to be a BayTrail based device if this deal would be true.
In both cases, there is no reason why Krzanich would have announced the BayTrail devices for november in case it would have been against their plans. Also, even if they'd have changed their plans just 1 month later, he has the responsibility towards the shareholders to explain the situation. He missed that opportunity at the investor meeting and conference call following that disappointment.
It is true his point was plain wrong but I still think it is good to have some guys from the evil side
I mean, do you guys really just want to hear how great Intel is and how good your investment is? If that's the case, why bother to read this stuff here on the forum? Fastpath and wthdik should continue trolling - welcome from my side. Maybe we should also have a deeper look at the arguments Qualcomm and ARM investors deliver.
By the way, bad things about Intel: 5 months passed already since Krzanich promised to have a majority of Android BayTrail tablets in the market/on shelves. Something he said in a conference call, not in some interview. That is really pi...ng me off. I really expect him to give an explanation in the next conference call about why that didn't happen and why we have to wait that long after the Windows release of BayTrail - the hardware is obviously not the issue here. Intel is loosing time with a not exactly killer product - that is harm to our investment and there needs to be an explanation.
Maybe it's Android 64, maybe it's driver issues with the graphics, maybe it's that noone wants to buy it but there needs to be an explanation.
I guess the analysts will once again instead ask questions about inventory - as if Intel was some supermarket and this would really matter to its business. These guys really need to do their homework ...
Yeah, I found that puzzling as well, I would have expected A7 to have a much shorter pipeline. Though, at a second thought, considering that they'll likely use the slower low power process edge (I suppose TSMC would tune their process for a customer like Apple) and that they go quite low with the voltage (something the 1,3 GHz tells me) it makes sense. All these measures typically affect timing in a negative way and you need the longer pipeline to compensate for that. Basically the same rules apply, whether you are developing a CPU for +3GHz at a power consuming high performance process or one for low power applications with a low power process - timing gets critical for both.
Kind of funny to see that the same design optimum for mobile and high performance seems to basically be the same core structure (oversimplified, I know). If Intel would just strip down Haswell - they would lead the market already. I really hope Goldmont ist basically that, otherwise it might be too late for that step.
Well, the same physics apply for Intel as for everyone else. No magic there either. Finfets may help, though. Again, the same measures that help at high performance seem to help at low power.
Interesting. Would someone do a comparison with Haswell? There's not really much that Apple seems to be able to do from here. Going wider doesn't make much sense and the core already is quite power hungry. They may be able to increase the clock a little bit at 20 nm, but 20 nm doesn't reduce the dynamic power that much, with higher leakage at the same time. There's not much room to raise the voltage I guess. I don't see how Apple will have twice the performance with A8 other than building a quadcore (which - did I mention it already? - is nonsense for mobile).
What I think is interesting is the huge (for mobile) L3 cache. I think that is what we've seen as "SRAM" on the A7 layout. All these micro benchmarks oh so popular in mobile will nicely fit in there, distorting the actual performance that can be achieved in real life quite a bit. It may make sense for graphic applications (i.e. games) where the bottleneck, DRAM bandwidth, is limiting. Something Intel did with Iris Pro (and which helps to speed up the CPU as well).
Anyway, Intel really needs to take care of its high end if they want to continue to charge a premium there. They need to advance their cores much faster than they did in the past (e.g. +10% Haswell). Maybe they should consider having a beefed up high end core which doesn't care about power consumption, e.g. for hardcore gamers, workstations etc. The Haswell successor could then be used for the tablet, Ultrabook, notebook and all-in-one segment.
As I said, everybody on this forum seems to just see the things that suits his investment best ...
What I was actually thinking about is the possibility that future Android releases might be 64 bit only, so no upgrades anymore for 32 bit devices (though those stagnating Android updates aren't needed anyway - Apple does a lot better than Google in my opinion).
I wasn't also thinking about upgrading memory but about future proveness of your device. If you buy a 4 GB RAM device today, chances are high it will still be enough in two years. Such a device will be quite expensive though. I have 1 GB RAM in my Android phone and yet it's running out of RAM ocassionally, closing apps in the background. I have a notebook running Windows 7 with 2 GB of RAM which almost never runs out of memory. So much about good enough in mobile ...
It's not so much about the Apps but about the OS (kernel, runtime and such). I guess they'll need a 32 bit comaptibility mode for Apps at the beginning. It's not a big issue for Java apps (well done Google), but there's native code as well, especially in games. By the way, Intel has a lot of know how in emulating those parts - maybe expertise needed by Google for a 64 bit port.
Would 8 wimpy cores where only one or two are used occasionally prove that? 64 bit may make people believe they'll get future OS upgrades, especially on a Nexus device. Maybe someone will release a 4 GB tablet or phone but memory prices grew significantly so that's not going to be cheap.
Now you take your "old smartphone", install the Netflix App, disable Wifi/use your mobile network and compare the speed of yours to that in the video and tell me it works just like the same.
Regarding 64 bit: Well, the video probably just showing that the 64 bit implementation is ready. Does 64 bit bring any advantage? No, it doesn't. Not for the user, no matter if it's with Intel, Qualcomm, ARM or Apple - just like 4 or 8 cores in mobile are complete nonsense.
Mediatek will stay at the sweet spot TSMC 28nm. No reason for them to pay for a high end process such as Intel's 14nm. Besides, since Intel isn't even able to integrate their own modem technology and Mediatek uses integrated comms as well, Intel couldn't fab for Mediatek right now, even if they would want to.
Intel's fabs are open for high end customers who pay a premium for the latest process tech. Altera is such a customer, Cisco may be one and Apple could be one as well. No Cheapos.