Apple Veterans Plotted to Kill PowerPC Long Before Jobs’ Return

Steven P. Jobs was unquestioningly the dictator at Apple, Inc. (AAPL) throughout his fruitful career.  During his time with the company he co-founded and twice served at CEO (or perhaps more times, if you count the medical leaves), he controlled every minor detail from the cafeteria layout at the Apple campus to the specs he wanted on each of his mobile electronics devices.  

And while recent revelations in his biography that he was willing to burn his company’s entire cash stockpile to try to legally destroy Google Inc.’s (GOOG) rival Android may be a tiny bit unsettling to shareholders; it’s hard to argue with the pure numbers — Mr. Jobs forged the most valuable tech company on Earth (in terms of market capitalization and profit).

I. Apple Long Planned to Ax RISC PowerPC Chips

But for all the products and transitions he spearheaded — the iMac, the iPod, iTunes, OS X, the iPhone, iOS, and the iPad — there appears to be one accomplishment that he reportedly was not the original driving force behind after all — Apple’s transition to Intel Corp. (INTC) CPUs in its personal computer lineup.

Apple’s PC market share lay decimated when Apple acquired NeXT just before Christmas back in December 1996. With NeXT came the return of Apple CEO and co-founder (who quit Apple in 1985 to found NeXT and serve as its CEO), Steve Jobs.

One key reason why Apple brought back Steven P. Jobs, reportedly, was to try to break free of the PowerPC architecture.  It worked, for better or worse. [Image Source: Risen Sources]
At the time of Mr. Jobs’ return Apple was still using the PowerPC architecture, a reduced instruction set computer (RISC) CPU design that Apple co-created with International Business Machines, Inc.’s (IBM) and Motorola, Inc. (who would eventually divide into Google’s recently acquired Motorola Mobility and Motorola Solutions, Inc. (MSI)).  The Apple-IBM-Motorola (AIM) alliance sounded like it would be a superstar.  But like most things Apple in the mid-1990s, it was struggling mightily.

Motorola, tasked with delivering the PowerPC designs in the 1990s, managed to get the G3 out the door in 1998.  And it was a solid design, if a bit late.  It went head to head with Intel Pentium II, and by many accounts it was winning.  Writes Low End Mac in a 1998 report:

The Pentium II is designed for a 66 MHz system bus and comes with a fixed 512 KB level 2 cache running at half CPU speed. That means the hotshot 333 MHz Pentium II is using a smaller, slower cache than the top end PPC 750 daughter cards.

To make matters worse, it’s also a less efficient design, still rooted in the CISC technology of the late 1970s. And the beast is still an energy hog, pulling 23.7W at 333 MHz, several times as much as the efficient RISC design of the PPC 750.

The latest Bytemark pegs the 333 MHz Pentium II at 4.7 for integer performance, 5.3 for floating point. In comparison, the discontinued PPC 604e at 350 MHz hits 10.3 on the integer test and 7.6 on floating point. When Byte tested a prototype 275 MHz G3 system, it produced a 9.4 integer benchmark, 6.1 floating point.
But according to a new Churchill Club talk by Larry Tesler, an Apple veteran from the 1990s, Apple was trying to ditch the PowerPC even back then.  Apple lured Mr. Tesler away from Xerox Corp.’s (XRX) PARC project to hep build its graphic user interface operating system.

When it comes to ditching the PowerPC, Mr. Tesler remarks:
It was actually one of the reasons that the company decided to acquire Next… We had actually tried a few years before to port the MacOS to Intel, but there was so much machine code still there, that to make it be able to run both, it was just really really hard. And so a number of the senior engineers and I got together and we recommended that first we modernize the operating system, and then we try to get it to run on Intel, initially by developing our own in-house operating system which turned out to be one of these projects that just grew and grew and never finished. And when we realized that wouldn’t work we realized we had to acquire an operating system, either BeOS or Next, and one of the plusses was once we had that we could have the option of making an intel machine.
The full roundtable, which talks about Steve Jobs’ contributions at Apple can be viewed below:

II. PowerPC — Did the Better Architecture Lose?

According to Rama Menon, a senior Intel engineer, the reasons PowerPC fell behind in the race was not because of the design itself, but rather the AIM alliance’s manufacturing capability and the smaller installed base.  The collapse of Apple computer sales meant there weren’t many PowerPC computers on the market, which meant there wasn’t much PowerPC software design going on — at least compared to x86.  

In 1998, the game was pretty even in terms of feature size.  Intel had just released the 7.5 million-transistor Deschutes on a 0.250 µm process, just months after the PowerPC G3 7xx series launched on a 0.260 µm process (late 1998).  But Intel killed the PowerPC backers in its ability to churn out close to 10 million CPUs a year.

Apple and its supporters long claimed PowerPC to be far faster than its Intel Pentium brethren.  But the lack of independent benchmarks made it unclear exactly how big PowerPC’s lead really was. [Image Source: Bryon Realey/Flickr]
In February 1999, Intel launched the Pentium III, a 9.5 million-transistor design that stuck at the 0.250 µm node.  Whatever problems there were between Apple and its fellow AIMers were about to get worse.  Motorola’s counterpunch to the P3 was PowerPC G4 (74xx), a 10.5 million-transistor chip built at 0.200 µm.  Motorola had apparently gone a bridge too far and was forced to back off its claim of 500 MHz, shaving its top speed to 450 MHz.  This was an embarrassment to Apple who had been boasting of its ultra fast “500 MHz” chips.

By February 2000, Motorola finally got its 500 MHz chips working, but Intel had set loose 733 MHz chips.  Then in May 2000, Intel aired a 180 nm chip clocked up to 1 GHz.  Of course, a 500 MHz PowerPC G4 was reportedly 1.67 times as fast as an 800 MHz chip in some benchmarks.  

But these kinds of claims were always a bit hard for consumers to digest, given that there was a relative lack of equivalent software — optimized for both the better selling CISC world and the RISC world — to pit the architectures against each other head to head.  Indeed, sites like AnandTech seldom compared PowerPC against P3s in benchmarks — or benchmarked PowerPC at all for that matter.

In 2001, with Motorola on the verge of bailing on PowerPC, IBM took over design duties, making the 64-bit PowerPC G5 (970), which aired in June 2003.  By 2005 Intel’s Pentium 4 had overtaken the lower clocked PowerPC G5 in key metrics like server performance.

III. The Endgame: Apple Goes x86

In June 2005 Steve Jobs aired Apple’s plans to kill PowerPC in its computers and transition to Intel chips.  By 2006, the golden era of x86 had kicked off at Apple.  The new chips brought with them improved compatibility and portability — after all, if you can get past the OS difference a Mac x86 is still and x86.

In August 2009 OS X 10.6 “Snow Leopard” shipped, becoming the first Apple operating system to drop PowerPC support.  The era of the PowerPC was officially over.  But if reports are to be believed, the beginning of the end began nearly a decade and a half prior, when a Steve Jobs-less Apple began plotting to revamp its MacOS and jump ship to x86.

With Snow Leopard (OS X 10.6), Apple officially gave PowerPC holdouts the boot from its latest shiny toys. [Image Source: Google Images] 
That could have been a huge business for IBM, given that Apple is today the third largest PC maker in the U.S.  But it’s hard to say whether it could have revived its market share to such healthy levels had it stuck with PowerPC and lacked base x86 application compatibility.

In doing so it might have killed a great thing.  PowerPC appeared far faster at processing and more power efficient than Intel, but the lack of a healthy quantity of unbiased benchmarks makes it hard to say if that advantage was real or but a mirage.

But for what it’s worth, even as Apple would put one veteran RISC architecture’s foot in the grave, it was boosting what was arguably a superior architecture — the Advanced RISC Machine (ARM, for short).  Today thanks to Apple’s early adoption in the iPod (and later Palm Inc. following in suit), ARM became the de facto standard for mobile devices, proudly carrying on the RISC tradition and enjoying a discrete chuckle at Intel’s struggling market entry attempts.