Doug Siebert (foo@bar.bar) on 1/24/12 wrote:
---------------------------
>According to Intel's stats, the reference platform uses a 5.55 Wh battery. The
>iPhone 4S uses a 5.3 Wh battery so those are pretty comparable.
>
>Other than standby time, which is 360 hours for the reference platform versus the
>iPhone's 200 hours, the other numbers show no advantage for Intel at all. The reference
>platform has one hour less talk time, two hours less video >watching time and one hour more browsing time.
>
>While the standby time is a big advantage, its the least >important one since few
>people leave their phone on standby for a week without >using it or charging it.

>There are three smartphone markets that anyone (other than RIM shareholders) think
>will matter in two years. iPhone, Android and Windows Phone. The first market
>is likely closed to Intel, the second will be the bulk of the market, and for the
>third it remains to be seen if it will amount to more than >Blackberry does today.

In the short run, I agree that iPhones will use Apple's internal designs. However, it's not clear that will last forever. It may be that 5-10 years from now, there is an opportunity for third party vendors.

>Intel needs a clear advantage if they want to do anything more than have a good
>chance of being the standard CPU for the Windows Phone products. If Intel ships
>22nm when ARM is at 28nm, and 14nm when ARM reaches 20nm, they must do better than
>match ARM as they appear to do here or they'll be left hoping that Windows Phone
>turns out to be more than just a niche market.

I think Intel's plan is to beat everyone on process technology, as you suggested. Remember it's not about ARM, it's about a combination of {TSMC, UMC, Samsung, GF} + {Qualcomm, TI, Samsung, Nvidia, Apple, Marvell, Broadcom}.

>Yes, I know Intel will make architectural improvements to complement their process
>improvements, but so will ARM.

I think the microarchitectural improvements will be bigger for Intel than ARM. As many people have accurately noted, Atom is not a fantastic core and in some ways is less advanced than the A9 (e.g. OOOE), while more advanced in others (e.g. multithreading, caches). The reason why Intel hasn't mucked around with the core much is that it's not the biggest lever for improvement. Obviously, full SoC integration was the biggest issue, along with all the platform changes it entails.

>The large number of ARM licensees mean there will
>be a much larger variety of ARM solutions to fit every >need, whereas Intel will
>have one architecture per generation and try to serve >everyone by varying the frequency
>and number of cores in the CPU and GPU.

If you look across the ARM licensees, there are ~2-3 custom designs, but most are implementations of the standard ARM references. I think it may prevent Intel from hitting the entire market, but frankly they are at the point where any marketshare is good. If they can only address 60% of the market, instead of 80%, that's not a big deal.

>This scenario is actually somewhat comparable
>to the wide variety of Android phones of all types and >pricepoints versus the iPhone
>trying to serve everyone by varying the age of the design >and the amount of onboard memory as their only segments.

I would think that price point is really the most important variable here.

David