ARM

Macintel: The End Is Nigh

When Apple announced its 64-bit A7 processor, I dismissed the speculation that this could lead to a switch away from Intel chips for the Macintosh line for a homegrown “desktop-class” chip. I might have been wrong.

“I don’t know exactly when, but sooner or later, Macs will run on Apple-designed ARM chips.” Thus spake Matt Richman in a 2011 blog post titled “Apple and ARM, Sitting in a Tree”. Richman explained why, after a complicated but ultimately successful switch from PowerPC chips to Intel processors in 2005, Apple will make a similar switch, this time to ARM-based descendants of the A4 chip designed by Apple and manufactured by Samsung.

Cost is the first reason invoked for the move to an An processor:

“Intel charges $378 for the i7 chip in the new high-end 15 inch MacBook Pro. They don’t say how much they charge for the i7 chip in the low-end 15 inch MacBook Pro, but it’s probably around $300. …When Apple puts ARM-based SoC’s in Macs, their costs will go down dramatically. ”

We all know why Intel has been able to command such high prices. Given two microprocessors with the same manufacturing cost, power dissipation, and computing power, but where one runs Windows and the other doesn’t, which chip will achieve the higher market price in the PC market? Thus, Intel runs the table, it tells clone makers which new x86 chips they’ll receive, when they’ll receive them, and, most important, how much they’ll cost. Intel’s margins depend on it.

ARM-based processors, on the other hand, are inherently simpler and therefore cost less to make. Prices are driven even lower because of the fierce competition in the world of mobile devices, where the Wintel monopoly doesn’t apply.

329_A7chip

Cost is the foremost consideration, but power dissipation runs a close second. The aging x86 architecture is beset by layers of architectural silt accreted from a succession of additions to the instruction set. Emerging media formats demand new extensions, while obsolete constructs must be maintained for the sake of Microsoft’s backward compatibility religion. (I’ll hasten to say this has been admirably successful for more than three decades. The x86 nickname used to designate Wintel chips originates from the 8086 processor introduced in 1978 – itself a backward-compatible extension of the 8088…)
Because of this excess baggage, an x86 chip needs more transistors than its ARM-based equivalent, and thus it consumes more power and must dissipate more heat.

Last but not least, Richman quotes Steve Jobs:

“I’ve always wanted to own and control the primary technology in everything we do.”

Apple’s leader has often been criticized for being too independent and controlling, for ignoring hard-earned industry wisdom. Recall how Apple’s decision to design its own processors was met with howls of protest, accusations of arrogance, and the usual predictions of doom.

Since then, the interest for another Grand Processor Switch has been alive and well. Googling “Mac running on ARM” gets you close to 10M results. (When you Bing the same query, you get 220M hits — 22x Google’s results. SEO experts are welcome to comment.)

Back to the future…

In September 2013, almost a year ago already, Apple introduced the 64-bit A7 processor that powers new iPhones and iPads. The usual suspects pooh-poohed Apple’s new homegrown CPU, and I indulged in a little fun skewering the microprocessor truthers: 64 bits. It’s Nothing. You Don’t Need It. And We’ll Have It In 6 Months. Towards the end of the article, unfortunately, I dismissed the speculation that Apple An processors would someday power the Mac. I cited iMacs and Mac Pros — the high end of the product line —as examples of what descendants of the A7 couldn’t power.

A friend set me straight.

In the first place, Apple’s drive to own “all layers of the stack” continues unabated years after Steve’s passing. As a recent example, Apple created its own Swift programming language that complements its Xcode IDE and Clang/LLVM compiler infrastructure. (For kremlinology’s sake I’ll point out that there is an official Apple Swift blog, a first in Apple 2.0 history if you exclude the Hot News section of the of apple.com site. Imagine what would happen if there was an App Store blog… But I digress.)

Secondly, the Mac line is suspended, literally, by the late delivery of Intel’s Broadwell x86 processors. (The delay stems from an ambitious move to a bleeding edge fabrication technology that shrinks the basic building block of a chip to 14 nanometers, down from 22 nanometers in today’s Haswell chips.) Of course, Apple and its An semiconductor vendor could encounter similar problems – but the company would have more visibility, more control of its own destiny.

Furthermore, it looks like I misspoke when I said an An chip couldn’t power a high-end Mac. True, the A7 is optimized for mobile devices: Battery-optimization, small memory footprint, smaller screen graphics than an iMac or a MacBook Pro with a Retina display. But having shown its muscle in designing a processor for the tight constraints of mobile devices, why would we think that the team that created the most advanced smartphone/tablet processor couldn’t now design a 3GHz A10 machine optimized for “desktop-class” (a term used by Apple’s Phil Schiller when introducing the A7) applications?

If we follow this line of reasoning, the advantages of ARM-based processors vs. x86 devices become even more compelling: lower cost, better power dissipation, natural integration with the rest of the machine. For years, Intel has argued that its superior semiconductor design and manufacturing technology would eventually overcome the complexity downsides of the x86 architecture. But that “eventually” is getting a bit stale. Other than a few showcase design wins that have never amounted to much in the real world, x86 devices continue to lose to ARM-derived SoC (System On a Chip) designs.

The Mac business is “only” $20B a year, while iPhones and iPad generate more than 5 times that. Still, $20B isn’t chump change (HP’s Personal Systems Group generates about $30B in revenue), and unit sales are up 18% in last June’s numbers vs. a year ago. Actually, Mac revenue ($5.5B) approaches the iPad’s flagging sales ($5.9B). Today, a 11” MacBook Air costs $899 while a 128Gb iPad Air goes for $799. What would happen to the cost, battery life, and size of an A10-powered MacBook Air? And so on for the rest of the Mac line.

By moving to ARM, Apple could continue to increase its PC market share and scoop much of the profits – it currently rakes in about half of the money made by PC makers. And it could do this while catering to its customers in the Affordable Luxury segment who like owning both an iPad and a Mac.

While this is entirely speculative, I wonder what Intel’s leadership thinks when contemplating a future where their most profitable PC maker goes native.

JLG@mondaynote.com

———-

Postscript: The masthead on Matt Richman’s blog tells us that he’s now an intern at Intel. After reading several of his posts questioning the company’s future, I can’t help but salute Intel management’s open mind and interest in tightly reasoned external viewpoints.

And if it surprises you that Richman is a “mere” intern, be aware that he was all of 16-years-old when he wrote the Apple and ARM post. Since then, his blog has treated us to an admirable series of articles on Intel, Samsung, Blackberry, Apple, Washington nonsense – and a nice Thank You to his parents.

 

Peak PC. Intel Fork.

 

Propelled by Moore’s Law and the Internet, PCs have enjoyed four decades of strong growth, defying many doomsday prophecies along the way. But, with microprocessor performance flattening out, the go-go years have come to an end. Intel, the emperor of PC processors, and a nobody in mobile devices needs to react.]

I’m suspicious of Peak <Anything> predictions. Some of us became aware of the notion of a resource zenith during the 1973 OPEC oil embargo, with its shocking images of cars lined up at gas stations (in America!):

Gas Lines Oil Embargo

This was Peak Oil, and it spelled doom to the auto industry.

We know what happened next: Cars improved in design and performance, manufacturers became more numerous. Looking at this bit of history through my geek glasses, I see three explanations for the rebound: computers, computers, and computers. Computer Assisted Design (CAD) made it easier to design new car models as variations on a platform; Volkswagen’s MQB is a good example. Massive computer systems were used to automate the assembly line and manage the supply chain. It didn’t take long for computers to work their way into the cars themselves, from the ECU under the hood to the processors that monitor the health of the vehicle and control the entertainment and navigation systems.

Since then, we’ve had repeated predictions of Peak Oil, only to be surprised by the news that the US will soon become a net oil exporter and, as Richard Muller points out in his must-read Physics for Future Presidents, we have more than a century of coal reserves. (Unfortunately, the book, by a bona fide, middle-of-the-road physicist, can’t promise us that physics will eventually push politics aside when considering the rise of CO2 in the atmosphere…)

I’ve heard similar End of The Go-Go Days predictions about personal computers since 1968 when my love affair with these machines started at HP France (I was lucky enough to be hired to launch their first desktop machine).

I heard the cry again in 1985 when I landed in Cupertino in time for the marked slowdown in Apple ][ sales. The never-before round of layoffs at Apple prompted young MBAs, freshly imported from Playtex and Pepsi, to intone the It’s All Commodities Now dirge. I interpreted the cry (undiplomatically -- I hadn’t yet learned to speak Californian) as a self-serving It’s All Marketing Now ploy. In the meantime, engineers ignored the hand-wringing, went back to work, and, once again, proved that the technology “mines” were far from exhausted.

In 1988, a Sun Microsystems executive charitably warned me: “PCs are driving towards the Grand Canyon at 100 mph!”.  A subscriber to Sun’s The Network Is The Computer gospel, the gent opined that heavy-duty computing tasks would be performed by muscular computers somewhere (anywhere) on the network. Desktop devices (he confusingly called them “servers” because they were to “serve” a windowing protocol, X11) would become commodities no more sophisticated or costly than a telephone. He had no answer for multimedia applications that require local processing of music, video, and graphics, nor could he account for current and imminent mobile devices. His view wasn’t entirely new. In 1965, Herb Grosch gave us his Law, which told us that bigger computers provide better economics; smaller machines are uneconomical.

And yet, personal computers flourished.

I have vivid memories of the joy of very early adopters, yours truly included. Personal computers are liberating in many ways.

First, they don’t belong to the institution, there’s no need for the intercession of a technopriest, I can lift my PC with my arms, my brains, and my credit card.

Second, and more deeply, the PC is a response to a frustration, to a sense of something amiss. One of mankind’s most important creations is the symbol, a sign without a pre-existing meaning: X as opposed to a drawing of a deer on a cave wall. Strung together, these symbols show formidable power. The expressive and manipulative power of symbol strings runs through the Song of Songs, Rumi’s incandescent poetry, Wall Street greed, and quantum physics.

But our central nervous system hasn’t kept up with our invention. We don’t memorize strings well, we struggle with long division, let alone extracting cubic roots in our heads.

The PC comes to the rescue, with its indefatigable ability to remember and combine symbol strings. Hence the partnership with an object that extends the reach of our minds and bodies.

Around 1994, the Internet came out of the university closet, gave the PC access to millions of servers around the world (thus fulfilling a necessary part of the Sun exec’s prophecy), and extended our grasp.

It’s been great and profitable fun.

But today, we once again hear Peak PC stories. Sales have gone flat, never to return:

PC shipments 2014-18 - PNG

This time, I’m inclined to agree.

Why?

Most evenings, my home-builder spouse and I take a walk around Palo Alto. Right now, this smallish university town is going through a building boom. Offices and three-layer retail + office + residence are going up all around University Avenue. Remodels and raze-and-build projects can be found in the more residential parts of town. No block is left unmolested.

I can’t help but marvel. None of this activity, none of Silicon Valley would exist without Moore’s Law, the promise made in 1965 that semiconductor performance would double every 18 months. And, for the better part of 40 years, it did - and rained money on the tech ecosystem, companies and people. PCs, servers, embedded electronics, giant network routers, cars...they’ve all been propelled because Moore’s Law has been upheld...until recently.

The 1977 Apple ][ had a 1MHz 8-bit processor. Today’s PCs and Mac’s reach 3.7GHz, but number that hasn’t changed in more than three years. This isn’t to say that Intel processors aren’t still improving, but the days when each new chip brought substantial increases in clock speed seem to be over.

One should never say never, but Moore’s Law is now bumping into the Laws of Physics. The energy needed to vibrate matter (electrons in our case) increases with frequency. The higher the clock frequency, the higher the power dissipation and the greater the heat that’s generated…and a PC can withstand only so much heat. Consider the cooling contraptions used by PC gamers when they push the performance envelope of their “rigs”:

EK-Thermosphere_right2_12001

To work around the physical limits, Intel and others resort to stratagems such as “multiple cores”, more processors on the same chip. But if too many computations need the result of the previous step before moving forward, it doesn’t matter how many cores you have. Markitects have an answer to that as well: “speculative branch execution”, the use of several processors to execute possible next steps. When the needed outcome appears, the “bad” branches are pruned and the process goes forward on the already-computed good branch. It makes for interesting technical papers, but it’s no substitute for a 8GHz clock speed.

If we need confirmation of the flattening out of microprocessor progress, we can turn to Intel and the delays in implementing its Broadwell chips. The move to a 14 nanometers  “geometry” — the term here denotes the size of a basic circuit building block — is proving more difficult than expected. And the design isn’t meant to yield faster processors, just less power-hungry ones (plus other goodies such as better multi-media processing).

One possible reaction to this state of affairs is to look at tablets as a new engine of growth. This is what Microsoft seems to be doing by promoting its Intel-inside Surface Pro 3 as a laptop replacement. But even if Microsoft tablets turn out to be every bit as good as Microsoft says they are, they aren’t immune to the flattening out of Intel processor performance. (I don’t have an opinion yet on the product — I tried to buy one but was told to wait till June 20th.)

Does this broaden the opening for ARM-based devices? Among their advantages is a cleaner architecture, one devoid of the layers of backwards compatibility silt x86 devices need. ARM derivaties need less circuitry for the same computing task and, as a result, dissipate less power. This is one of the key reasons for their dominance in the battery-powered world of mobile devices. (The other is the customization and integration flexibility provided by the ARM ecosystem.) But today’s ARM derivatives run at lower speeds (a little above 1GHz for some) than Intel chips. Running at higher speeds will challenge them to do so without hurting battery life and having to add the fan that Microsoft tablets need.

With no room to grow, PC players exit the game. Sony just did. Dell took itself private and is going through the surgery and financial bleeding a company can’t withstand in public. Hewlett-Packard, once the leading PC maker, now trails Lenovo. With no sign of turning its PC business around, HP will soon find itself in an untenable position.

Intel doesn’t have the luxury of leaving their game — they only have one. But I can’t imagine that Brian Krzanich, Intel’s new CEO, will look at Peak PC and be content with the prospect of increasingly difficult x86 iterations. There have been many discussions of Intel finally taking the plunge and becoming a “foundry” for someone else’s ARM-based SoC (System On a Chip) designs instead of owning x86 design and manufacturing decisions. Peak PC will force Intel CEO’s hand.

JLG@mondaynote.com