Intel’s Third Embarrassment of the Last Decade - Chapter I

If anybody remembers Intel’s Atom launch, you’ll also remember how the company hoped that their chips will end up in everything from vacuum cleaners to fridges, cars and phones. With Intel’s “bribe” funds, incomparable R & D funds and unending monopoly of the x86 computing world, this was probably the only domain where everybody “knew” the giant won’t be able to succeed in if they keep on holding to the x86 ISA. The Medfield launch two weeks ago was a total success and we were all surprised.

The x86 ISA has so much overhead that everybody thought and often stated that there will never be an x86 CPU cool enough to be used on large scale in phones.

The CPU giant bragged about powering phones with x86 CPUs on yearly basis and everybody was quite used to see reality prove them wrong every time.

So imagine the surprise on everybody’s face when Medfield proved to be exactly what Intel promised it to be and even better. Don’t get me wrong, the expectation were that Medfield would fail. 

Most believed it would power a smartphone with the lowest battery life among the rest of the smartphones and that the performance would be incomparable with its ARM counterparts.

The thing is that, even if the battery life manifested by a Medfield powered phone was only average, it was average among the best of the smartphones out there. Rolling around with the best from the first product is quite an achievement for Intel.

The performance was expected to be especially low as x86 has a lot of legacy overhead and most phone software is build and optimized for ARM. 

Seeing Intel’s coolest Atom processor ace some benchmarks and give a decent showing in other was amazing.

We even started thinking that if they’ve cracked into the phone market, in the next decade we would see Intel’s funds “working” the “free” market and many players in the phone market heading towards a “better” place. 

Our future was set to be one where Intel’s mobile CPUs had no real competition and were costing 300 dollars a piece in 1500 dollars smarterphones.

The surprise of Medfield’s performance was so great that most of us never got a chance to lean back few days and see the big picture.

As life beats the movies, we were handed the end of the plot just a week later once Samsung’s powerful Galaxy S3 launched after the perfectly designed and superlative HTC One XL (AT & T Qualcomm version).

The reality was that Intel’s Medfield flagship, as shown in Anand’s exquisite review of Intel’s first foray into the phone market, was battling phones and platforms that were quite old in the market and some were on their way out.

Sure the ultra optimized Atom, using absolutely the best bulk 32 nm manufacturing process available today (coming from Intel’s own FABs) , was able to hang out with last year’s 45/40 nm ARM processors.

But this was Intel’s best Atom development put against ARM’s Cortex A9 architecture launched almost six years ago, back in 2007.

Intel practically had a chance to look at ARM’s 2007 Cortex A9, even license it if they’d wanted, and then, in 2008 launch their own low power design named Atom.

We’re not trying for a single fraction of a second to suggest Intel took anything from ARM’s A9. If it would have, it would have made Atom better than the “complete scam” it was at the time.

With that put aside, we must mention that Intel was and still is an ARM licensee and that they’ve even had their own ARM division called Xscale that was later sold to Marvell.

The reason Intel stopped its own successful ARM development and sold its division to Marvell we’ll reveal in our next Intel Atom related analysis.

Being able to mingle with the 5 year old Cortex A9 champs and taking some surprising swings using your 32 nm gloves and somebody’s else GPU was quite an achievement for Intel.

Unfortunately for the American/Israeli company called Intel, the party lasted only one week.

This last week, ARM’s A15 stepped into the ring. 

It was dressed as the HTC One XL knight and the role fit perfectly. Built in a 28 nm bulk process, Qualcomm’s Krait is one step ahead of Intel’s Medfield where FABbing is concerned.

This is reality and is actually the opposite of what Wikipedia’s “Atom (system on chip)” entry says about Medfield :

“Even though the architecture by its own most likely still is a bit less power efficient than ARMs designs, the advantage Intel has is their own manufacturing capabilities which makes it possible to stay one step ahead in semiconductor technology, currently using a 32 nm High-K/metal gate process whereas most competitors SoCs as of Q1 2012 are built with a 40/45 nm process by for example TSMC.”

Is that so? Not really! 

As we can easily see, even Intel’s Clover Trail is still built in 32 nm. Clover Trail is a bit faster on the frequency side as it works at 1800 MHz compared with Medfield’s 1600 MHz and brings a doubled GPU (don’t worry, this GPU is not Intel’s merit either).

ARM’s A15 on the other hand is built using TSMC’s 28 nm process and TSMC is investing heavily into doubling its 28 nm capacity and even in speeding up the 20 nm process.

Intel’s next Atom will consume even more power than the current one. It’s even targeted for bigger devices such as tablets.

So, looking at Clover Trail like looking at a mobster with a bigger gun (to be read GPU) is not going to save Intel from the smartphone disaster they’ll experience because of A15.

One thing’s for certain, even if Medfield is able to swing with Tegra 3 and dual core A9 CPUs, it can only hope to receive just 1% of the smarphone profits this next year.

More on what the Cortex A15 and ARM means for Intel’s ill fated attempt to take ever the mobile phone market in the following chapter.

Posted in: Hardware, News