ARM's INFERIOR chips at a higher price CANNOT win in the marketplace over the long run. PERIOD.
ARM is a severely hobbled architecture. It has now become a power hog with low performance with a high price tag compared to X86. Kludges such as Big little etc.. are done to compensate for its inferiority.
In a few years, ARM will be relegated back to where it belongs, in the embedded market.
We're entering an era where computers are no longer designed bottom-up, starting from the microprocessor that is going to power them but top-down, starting from how the computer is going to be useful to its user and working down to the choice of chip from there. You can read more about this trend in this rather old article:
Firstly the 30-40% increase is either from 3-way OR OoO but having both would compound the improvement i.e. 70-90% which is why the Atom design still has great performance potential. You can mitigate power increases by say shortening the pipeline as ARM did going from A8 to A7. Also the new 22nm trigate process will help in reducing power.
Intel doesn't have the experience in low power design. They're making things up as they go and then seeing how well they did. This why it has taken them 6 years to come this far and they still don't really know how far behind they are. They're pretty much blowing in the wind.
They are not behind at all. In the same exact phone, Medfield beats one of the most power efficient ARM chips, Snapdragon S4 Krait, in both performance and power efficiency. You can pretend otherwise but these facts are out there now and no amount of hand waving from you can change that.
Intel is beating 3-way out-of-order ARM designs like Qualcomm S4 in both performance and power efficiency with a stripped down slugged 2-way in-order x86 design like Atom. The architectural equivalent of Atom in the ARM world is the ARM A8, 2-way issue, in-order instruction processing, pipeline 10, and that just gets slaughtered by Atom so obviously x86 has a clear innate performance superiority under the same design conditions.
Just for clarity:
As for architectural arguments, you are right about the details (BTW, the A9 is only partial OO, the S4 more so, real deal is with the A15) but architectural differences/changes are really about transistor budgets driven by power and cost constraints used for the design and how many of those transistors light up as the ISA flows through them. If we look at the A8/A7 (2 way, in order) they are tiny compared to Atom, and in the A7's case, it would give it a run for it's money in some specific like-for-like benchmarks.. Even the larger A9 is much smaller than Atom. How many A9's can fit into the same die size as an Atom CPU core for example?
As we're talking about different ISA's it's not as straightforward as using high level architectural constructs as a means of comparison. A simple example is branch prediction, relatively simple on x86, but difficult on ARM (due to conditionals, nearly every instruction can branch). Another example is hyper threading, perceived as an advantage to Intel. Yet it in new Atom designs, it gets dropped for actual cores.
Ah the blue red monk arrives to fight for the ARM corner, someone at least I can have a decent conversation with ! lol
Sure, the ARM chips are smaller but
a) they never include their L2 cache in die size measurements which exaggerates the difference
b) you are still limited to only have four low performance or two medium performance or half a high performance ARM core vs an Atom core (with hyperthreading) due to power considerations irrespective of die sizes.
High level architectural constraints are still important because they highlight what is left architecturally as future POTENTIAL. Atom can gain in the future from both out of order processing and 3-way issue and gain 30-40% performance increase from each step. A15 is already there and frankly compared to past x86 3-way OoO processors it is disappointing.
An Intel employee (biased I know) reckons Bay Trail will be the Conroe of the tablet world. For those who don't know Conroe was the desktop Core 2 Duo and basically killed AMD in 2006 as a profitable company with a future ...
Is it due to interpreted code ?? Or the ARM small register set for interpreted code...
ARM's best effort so far the A15 is bested by Clovertrail and Baytrail in power and performance. End of Story!!. All the low hanging fruit and some high hanging fruit has been plucked. ARM just cannot keep up against X86.
Every 10-15 years, the same CISC vs RISC arguments crop up, Intel kicks butt and the RISC people go away. RISC couldn't even win in the Server market which it owned, where X86 compatibility was non existant.
In a few years there won't be any lunch left for intc. There are too many companies invested in arm there is no way intc can win on all the fronts it's trying to fight. Once AMD finally folds the Intel will be the sole company supporting x86 while arm has numerous - that's a lot of r &d to contend with.
As for your technical assessment - you are clueless. If you want an examples of kludges take a close look at the evolution of the x86 - everything from its addressing schema to instruction sets.
'There are too many companies invested in arm there is no way intc can win on all the fronts it's trying to fight. Once AMD finally folds the Intel will be the sole company supporting x86 while arm has numerous - that's a lot of r &d to contend with.'
Yawn this is not a new situation Intel finds itself in, having beaten all the RISCs silly for the last 2-3 decades (DEC Alpha, IBM Power, SUN Sparc, HP PA-Risc, SGI Mips etc) in both client and server and all the R&D put in them. For some strange reason a few large OEMs have decided in their very finite wisdom to pick the weakest and barest featured RISC to all have another go e.g. ARM. Having seen their efforts so far and their future roadmaps they have come up short ... again. Like Otellini said, micro-architecture is hard and Intel have the right architecture, CISC e.g. x86, and the right experience and patents to continue winning pretty easily.
There are too many companies invested in arm there is no way intc can win on all the fronts it's trying to fight
ST is pulling out
Freescale is backing off
Renesas cutting anothr 3000 jobs
NVDA not happy with TSMC 20nm
Marvell seeing negative ROI @ 20nm
BRCM backing off sticking with 28nm
You're wrong. Intel's going to take quite a bit of the embedded market, too :-)
Depends on your definition of embedded. Deeply embedded, no (do a search for TSMC's deal with Intel to port Atom to it's platform/process, and they search for why it never got anywhere). Intel are still using ARM in it's deep environments, even that 22nm LTE modem will use an ARM CPU, not x86.
Intel will never take the embedded market. Can u give me one embedded example where it would be cost effective to use an x86?!? Anybody who'd use an x86 for a true embedded solution is nuts. I can use an nxp 32 bit arm proc for like $2 bucks.