The secret to ARMH is essentially the media and investors believing that there is something magical about ARM based chips that makes it impossible for Intel to compete. Just ask any of the people in the media who claim that ARM based chips are more efficient than Intel's legacy x86 chips blah blah. These people haven't the faintest idea that this is incredibly ironic as Silvermont was so damn focused on maximizing performance per watt and figured the ARM guys were doing the same. Instead, they just raised the power envelopes to achieve higher performance.
Long term I believe Intel and Qualcomm will be unbeatable in modems leaving them as the last two vendors standing. I also think that if Intel can demonstrate real leadership in phone SOCs from a phone perspective even QCOM will have a tough time matching the performance and power of Intel's finest without having a wildly higher cost structure.
Apple is the open question as I believe long term modem integration is necessary. When will they buy a modem team? I believe Apple May buy out Broadcom's entire wireless op long term as QCOM and INTC integrate all of the connectivity functionality into their chips leaving Broadcom's now money losing wireless biz in deep trouble. They're probably just waiting for an opportune time to pounce...
You made an interesting comment in one of your posts that Intel dropped on unrealistic analyst expectations, but I humbly submit to you that it was pumped up on high analyst expectations. The word was out that PCs did better than expected (and Intel continued to take market share from AMD), so the analysts plugged that into their models that assumed as-expected DCG and better-than-expected PCCG. Unfortunately, DCG came in weak which meant that the ~$27, which was predicated on those two factors, could not hold.
You were correct in calling the technical breakout, but that breakout - as always - was based on a market belief that PCs had done better than expected AND that DCG would come in-line. Unfortunately, as a programmer, you know that for the uptrend to have held, BOTH conditions would have to hold true for the thesis to hold true. They didn't, thus the breakout could not be sustained.
Technicals tell you how the market is interpreting current company/macro fundamentals, IMO.
WW probably deleted it. If Yahoo! deleted it, it was certainly not at my request. Once again, you accuse me unduly of reporting posts that I did not. I did report abuse on the psycho alias with the disgusting wallis- insulting alias. Nobody deserves that kind of treatment.
I can take the criticism of my work, but your criticisms have largely turned into attacks on me and my motives.
"I also in that thread said use Moorefield as the premium android part because of its gpu. It is clear I copied nothing from your article as you falsely claimed."
I don't actually think you "copy" anything I write...I actually fully believe in your ability to perform this type of analysis and actually enjoy your posts. This is why I find it unfortunate that you continue to attack me as I have not any ill-will towards you.
"Red herring, just an implementation feature. There is a single IBM CISC mainframe instruction that breaks down into thousands of micro-ops, try coding that in RISC lol !"
What's the latency of that instruction? Also, what instruction is it? I'm curious :-)
"That Ihub thread was interesting as basically it showed I had to convince you that Merrifield/Moorefield were better suited for Android than Bay-T and now you claim I got the idea off you lol ! Seriously it's been obvious to me for sometime that you do not remember importantly what you say and what is said back to you."
You didn't need to convince me as we were both in agreement that Merrifield/Moorefield had a far superior GPU - this was obvious from day 1 from the Apple A7 implementation of G6430. MEF and MOF should both have meaningfully higher performance.
My only question (and it is still a valid point), is why Intel claimed that Bay Trail-T would be the big runner in tablets rather than Merrifield in the low end and Moorefield at the high end for Android. I think that if we do see BYT-T dominate Intel's Android stack, it would be due largely to the fact that GenX probably has more effective media/2D engines (QuickSync comes to mind). Intel may also want to encourage developers to optimize for the GenX.
I also found the data-sheet on MEF and it says the memory interface is 2x32 bit 533MT/s. If MOF is 2x32bit 800MT/s (as suggested by the release slides), then BYT-T will have more memory bandwidth in its top configuration than MOF (although this seems more academic).
My point still stands: that you accuse me of not having an original thought when you clearly know better is sad and unfortunately reflects the fact that you simply dislike me.
You suggested that Merrifield would be suitable for a tablet chip a while back, but I don't see how this really proves your claim. Intel itself had said at Computex 2013 that Merrifield would be for both tablets and phones.
Indeed, way back in September when Intel announced a 2 core BYT (i.e. defective dies), I said that Merrifield made more sense and that using disabled BYT wouldn't really make sense in the time-frame that Digitimes claimed that Intel would launch the dual core BYTs. Your response at the time was that they'd use BYT until they got the cost reduced MEF chip out for this space...my argument was that MEF was supposed to be a Q1 launch. So, really, I think your claims that I don't have an original thought is perhaps misguided.
On your point re: X86 v.s. ARM, it's interesting that the "big core" cracks x86 instructions into micro-ops while SLM just runs x86 straight through the pipeline. On one hand you're claiming that by virtue of more powerful X86 instructions Intel can reach higher performance than ARM, but the highest performance X86 cores are effectively RISC internally.
I think where you may have some validity to your point is that the X86 SIMD extensions found in something like Haswell are far more powerful than NEON. If Atom begins to more aggressively adopt these extensions *and* if ARM doesn't revise its ISA as quickly as Intel does, then the ISA argument on your part has merit. But as it stands, ARMv8 mobile chips and the Atom appear roughly even.
Let's also not forget that when I went ahead and did an analysis piece on why Moorefield and not BYT-T would be more suitable for a Nexus 8 given the time-frames and performance characteristics and it ended up being parroted by a number of other sites, you actually ended up making a post on iHub claiming the very things that I did in my piece. While it doesn't take a genius to actually claim that Moorefield is a better fit for a potential N8 than a Bay Trail-T, the fact is that I had worked through and published this idea well before you made your post.
I'm also going to take the time to disagree with you now on your claim that because ARM has played its "3-issue" and "6-issue" cards that X86 has more headroom than ARM. X86 may indeed have more headroom but this is because as the processors get more complex, Intel can draw upon the "greatest hits" from its big cores to shape the development of its small cores. Pick the most power efficient/sexy features and build a low power core out of them. But it has zilch to do with issue-width. Silvermont and Saltwell are both 2-issue, but they are worlds apart in perf/clock and perf/watt even normalized for process.
The ARM guys can improve their designs, too, as can Intel - I am arguing that the instruction set architecture isn't a big deal either way. It's about making the smart micro-architectural decisions, the quality of the physical design teams, and even the quality of the compilers and how easy Intel makes it for developers to optimize for their chips.
Let's also not forget that you have continually ignored the fact that there's far more to an SoC than just the CPU. and GPU. Not saying Intel can't do world class ISPs and the like, but it does need to work with the device vendors more closely to know exactly what features to put in (and Intel acknowledged that they are doing this with Broxton).
"I don't mind really but still how much original analysis is actually going on in his mind ?"
Quite a bit more than you give me credit for. For example, when I told you that the low end smartphone market required integration to succeed, you continued to go on about how a discrete modem + apps processor would work just fine. Then, at the Investor Meeting, Intel announces SoFIA. At further conferences, BK specifically points out that Intel's customers have been asking for integration as the #1 thing they want to see from Intel.
Then you took the Merrifield WebXPRT #'s at face value without even asking what browser it was used on - you just assumed that Snapdragon 800 was a "Snapless Dragon" without digging further. Turns out Intel was using "stock browsers" for each device - so the #$%$ Samsung browser on the S4 and the highly tuned Webkit based one for the Intel reference design. 3D Mark Physics says S800 is not snapless ;-)
Of course, you then tried to claim that Medfield/CLT+ would be cheap ARM killers when I told you that these SoCs were missing a lot of key functionality/integration that would prohibit their use as long-term weapons against ARM. You still claim that they are even when Intel is rushing to get SoFIA out the door to address these markets. Integration is he key to the majority of the smartphone market.
Further, you've claimed that GPU performance doesn't matter despite a major part of tablet usage is indeed in gaming. Apple spends a lot of die area on GPU for a reason, and Samsung obviously pushes QCOM to beef up the graphics as much as possible.
Finally you claim that none of the major handset vendors want to work with Intel as they want to keep Intel out. This directly contradicts Intel's own statements at the Investor Meeting that it would be focusing on the half-dozen smartphone vendors and no longer going after the small players. Do you think Intel has no shot if it has a world-class product?
Hmm...the tricky part is that Intel's modem doesn't support CDMA, so Intel would basically need to win an AT&T/T-Mobile only SKU of a major phone. Not that this is impossible (with the right contra-revenue/NRE support, Intel could do it), but it'll make it tricky.
Intel has deals with Lenovo/Motorola, Dell, and ASUS. I doubt ASUS will attack the US market, nor do I expect Lenovo under its brand this year, but Motorola could as could Dell if it decides it wants to be in phones again.
My best guess? A GSM version of the Moto X 2 would be the design to watch. If Intel can win that, it'll finally prove itself in the US in phones.
You know that the Motley Fool report they're citing is an article of mine...right? ;-)
"This is going to be a generational war as was the original x86-Risc war of the 80s/90s/00s, the one that left Risc to scurry up and down the food chain to enterprise servers and micro-controllers to escape x86."
Except there's no guarantee that Intel will be the one who wins. For all you know, X86 ends up being pushed further and further up as ARM continues to march into the mainstream.
The fact that any OEM can customize Android and the fact that anybody can have an ARM license as long as they pay for it is a problem not easily overcome by Intel. It will take many generations of extremely good execution, outgunning all of the other ARM licensees by putting that process advantage to work.
So far, Intel has been unable to execute and QCOM and MDTK are having a field day at 28nm while Intel still has yet to see Merrifield in the hands of consumers.
"Those $7 A7s will have bigger things to worry about like how do they compete against 73 sq mm Merrifield when it is eventually used as a performance/price weapon against them."
Nonsense. If Merrifield were really a great perf/price weapon against low end ARM, there would be no need for SoFIA on TSMC 28nm.
theblueredmonk has a point. The x86 decode block alone would probably blow Intel's transistor budget in trying to compete against ARM in this space.
If Intel is serious, it will finish the job that it started, take a bigger equity stake in ImgTec (which owns MIPS) and build MIPS micro-controllers for IoT. This hurts ARM while at the same time empowering a company in which Intel is already majority shareholder.
I really don't think you understand the "IoT" market. It's just a fancy word for micro-controllers and, frankly, this market isn't one that plays to Intel's strengths. Intel is in the business of computing and its process lead and vast experience in high performance designs is how it differentiates in these performance sensitive markets.
With "IoT" we are talking about micro-controllers where CPU performance isn't the star of the show, and we are also talking about a market where the requirements for each device are so different that one chip-maker can't "rule them all.
This is where ARM's business model (i.e. license various flavors of low power IP) comes in real handy. I think that these IoT oriented chips will be made by either the item vendors themselves or they will be made by a bunch of smaller companies. I doubt if the market for these chips is in aggregate worth Intel's time.
To put things in perspective: the mobile SoC market is worth about $30 billion. Atmel, Microchip, TI (which makes a lot more than micro-controllers) don't even have revenues that add up to $20B
Just something to think about ideal ;-)
Eh, while there are definitely reasons to be irritated with Intel's mobile progress, his assertion that Intel has no brand cachet is completely off-base. Intel was the 9th most valuable brand in the world according to Interbrand.
That being said, I do think Intel has no business pursuing wearable devices. The mobile chip market is so large and Intel has so little of it that wearables can wait until the core business is shored up.