Feb 8, 2006

Why Apple Really Ditched PowerPC

Apple wants to make their switch to Intel chips seem like a no-brainer, but the reality of it was a lot more complicated than just faster chips for Macs. Apple's claims of their Intel systems being "4-5x faster" than their PowerPC systems is a little much to swallow, especially with Intel Macs landing in users' hands and failing to live up to the hype. So if these Intel chips aren't really that much faster than the G5, why did Apple make the switch? The answer to this question is a lot more interesting than what Apple's telling you.

But First, Some Background

The beginnings of Apple's departure from PowerPC began in 2000. After introducing the G4, Apple found itself mired in the 500 MHz Fiasco, where Motorola wasn't able to push the G4 faster than 500 MHz for over a year. To alleviate this, Apple added a second processor to the Power Mac line but soon found itself with a second problem: Motorola's fabs couldn't keep up with demand. Needing to get out from between these two problems, Apple brought in IBM to help fab the G4.

ProcessorYearMax Clock
PowerPC G41999450 MHz
2000500 MHz
2001866 MHz
20021,250 MHz
20031,420 MHz
20041,500 MHz
20051,670 MHz
20061,670 MHz

Figure 1 Maximum PowerPC G4 clock-rate by year (based on Apple product releases).

In January '01, after months of intense development, the new PowerPC G4 debuted at speeds of 666 and 733 MHz. The enhanced G4 was able to end the 500 MHz Fiasco, though updates would once again be meager. The new G4 crawled to 1.42 GHz in the Power Mac and would make it to 1.67 GHz in the PowerBook, but it wasn't enough. Apple began looking for its next-generation processor, this time seriously considering Intel. IBM, however, had other ideas.

While all this was going on, IBM had released the Power4, a significant revision of its mainframe core. In fact, it was the fastest chip on the market at its release, beating out Intel's Itanium, Hewlett-Packard's PA-RISC, and even DEC's Alpha. That and the fact that IBM had proven itself capable of assisting Motorola at the same time piqued Apple's interest, and after weighing its other options — a tumultuous switch to Intel or a slow death with Motorola — Apple had signed up with IBM for a Power4-based PowerPC chip.

In June '03, at MacWorld New York, Apple announced its long-awaited successor to the G4: The Power Mac G5, using IBM's PowerPC 970. The 970 was a lite version of IBM's Power4 chip and ran up to 2 GHz on a 1 GHz bus. It also had VMX, IBM's implementation of the Velocity Engine, and support for much faster RAM than the G4 did. On top of all this, it was fully 64-bit but could run 32-bit code natively. In short, the G5 blew the G4 away in every way possible. To top it all off, Steve Jobs promised another thousand megahertz in just a year.

Reaction to the G5 was good and users scrambled to get their hands on the new systems. Mac OS X went through several optimizations for the new architecture and all seemed well at Apple's high end. But then things slowly started turning dark. At MacWorld '03, instead of delivering his promised 3 GHz systems, Steve Jobs's announcement topped out at 2.5 GHz, and the G5 was getting nowhere near the PowerBook line, running too hot and drawing far too much power. The G5 was failing to meet its expectations.

ProcessorYearMax Clock
PowerPC G520022.0 GHz
20032.5 GHz
20042.7 GHz
20052.5 GHz

Figure 2 Maximum PowerPC G5 clock-rate by year (based on Apple product releases).

At MacWorld '05, Apple once again upgraded its Power Mac line, this time pushing the chip just 200 MHz faster than it had been a year ago, the elusive 3 GHz still nowhere in sight two years after it was promised. The G5 was still nowhere near getting into the PowerBook, and vague rumors of a new PowerPC core for portables cropped up. The G4 had practically stalled out at this point and the G5 was looking more like a stop-gap and less like a real solution to Apple's chip problems.

Apple had stuck to offering dual processors to keep competitive with Intel and AMD; now, as this tactic was running out of steam, IBM released dual-core versions of the G5 in order to offer Apple the benefits of symmetric multi-processing without the engineering overhead for a second chip. The result was the PowerPC 970MP, which per core ran slower than previous versions but, as a consolation, offered twice the cache. At this point, it became evident that IBM had seriously failed to meet its goals for the G5.

Outnumbered And Outgunned — Again

Apple was now in the same position it had been in five years earlier, with a processor that was falling behind and needing a new chip supplier desperately. Since moving away from the G4, Motorola had spun its chip division off into Freescale, with a G4 whose clock hadn't budged in years and shadowy promises of 64-bit dual-core chips that weren't even ready for sampling. Apple's portable line had been hit the hardest, standing still with the G4 while Pentium derivatives cleaned house. Apple was in dire straits.

But Apple had a secret — for the last five years, Mac OS X had never lost its ability to run on Intel chips, even though Apple had stopped releasing Intel versions of it after Rhapsody DR2. OpenStep had run on Intel processors quite nicely ever since the days of the 486, as well as on Solaris and Windows NT. Mac OS X was truly a cross-platform beast. With their platform transition experience, and a stable API and mature development tools, Apple took a long hard look at Intel again.

With its PowerPC experience souring its taste for the architecture, and with no other PowerPC suppliers to turn to, Apple's choice of Intel was natural; it wasn't just about trustworthy roadmaps and performance-per-watt. But the degree of failure on Motorola and IBM's parts begs questioning. How could such large tech companies, known for their micro-architecture businesses, have failed Apple so grievously? After such a powerful start in the mid-Nineties, what could have possibly gone wrong?

The situation with Motorola is simple: Motorola is focused on the embedded space, which doesn't need faster processors every quarter and new cores every couple years. And compared to their embedded customers, Apple was just not big business for Motorola. The G4 had more use in the networking and industrial sectors than it did in Apple, and Motorola didn't want to invest themselves in making updates to their chip that only one customer needed. Motorola was being conservative, which is what they do best.

The last time Motorola created any new cores was with IBM in 1995, when they introduced the PowerPC 602, 603, 604, and 620 parts. By '97 development had shifted to a third major chip, which Apple called the G3. At this point IBM went back to their own chips and Motorola, left alone to create a new core for Apple, simply updated the G3 with AltiVec and a better complex math unit to create the G4. In other words, Motorola has been stretching the same technology for over ten years.

Too Busy For Apple

But what about IBM? Isn't this the same company who created the PowerPC architecture to begin with? Don't they have technology that AMD and Intel covet? They were the proactive force behind PowerPC in the Nineties, so what happened to them and their program with Apple? Unfortunately, this question is a little more difficult to answer than Motorola's, and it has a lot to do with IBM as an innovator. It starts back in '99 with Nintendo's request to create a small, cool, fast chip to run its next-generation game system with.

IBM created for Nintendo the Gekko, a PowerPC G3 derivative, and eventually sold over 20 million of them. Smelling success, IBM started focusing on the highly-specialized embedded applications and courted other console makers. Their campaign paid off and Power was to be at the center of the Microsoft Xbox 360, Nintendo Revolution, and Sony PlayStation 3 development. IBM had finally realized its niche in consoles after failing to achieve this in the workstation marketplace for years.

ProcessorCoresClock
PowerPC 970FX12.7 GHz
PowerPC 970MP22.5 GHz
Xenon33.2 GHz

Figure 3 PowerPC G5 vs. Xenon.

Microsoft's Xenon processor was to take the most development resources from IBM to complete, as the chip for the Nintendo Revolution was not much different from Gekko and the Cell was a joint effort between IBM, Sony, and Toshiba. Microsoft insisted on a PowerPC G5-like chip with high bandwidth and multiple cores. Spread between its own chip needs as well as Apple's, Nintendo's, Sony's, and Microsoft's, engineering resources were run thin. And guess who was the first to feel it?

Apple, that's who, with its measly million or so G5s per quarter. It's easy to see how Microsoft's Xenon project sabotaged Apple's G5 when you stack the numbers up. At best, the G5 ran at 2.7 GHz but actually only hit 2.5 GHz when running dual cores; Microsoft's Xenon, however, packed three cores running at 3.2 GHz. Finishing the Xenon was IBM's priority and Apple suffered for it. Had IBM not been pressed away from the G5, we might see it running at 3.2 GHz and packing two, three, and possibly four cores a piece.

Better Off Without Power

After looking at Motorola and IBM and their chip businesses, it becomes clear that Apple was a victim of apathy, and that Apple really needed a chip company whose business was desktops, portables, workstations, and servers — not routers or mainframes or game consoles. These criteria basically left one choice for Apple, and there was no one to dissuade them this time. Apple announced their switch to Intel processors in June '05 and just recently announced and released their first such systems to wide acclaim.

The rest, as they say, will be history. The Intel transition will be the smoothest one Apple has ever gone through, much smoother than the 68k-to-PowerPC switch and a lot quicker than the Mac OS X migration. With their own development tools, and Tiger presenting a mature, stable API, the cross-grade to Intel chips is almost seamless. And Apple will never again find itself embarrassingly behind its rivals for clock-speed, or number of cores, or any other feature that becomes important to CPUs. It's obvious Apple wants to complete the move as quickly as possible and get on with things, which is why they're six months ahead of the Intel schedule.

When all is said and done, Apple knows one thing for sure: The sooner forgotten PowerPC is, the better.

7 comments:

  1. Thank you for this article. It was extremely informative and very well written.

    ReplyDelete
  2. IMHO, G4 was one of the best things in the history of personal computing... just compare it to the Pentium 4 joke :) Apple has just been pushed into the corner by Motorola/Freescale and IBM and I'll always regret that the G5 was not a successor to the marvellous G4 but rather a shoddy offshot of POWER4.....

    I still have a 733MHz G4 Quicksilver and it is a magnificent beast.

    ReplyDelete
  3. IBM didn't take a second look at Apple largely because Apple wanted IBM to do all the chip work for just them. If they had ponied up the dollars and/or manpower, things might be a lot different.

    As it stands, it's PowerPC for me all the way. I will never go back to Intel.

    ReplyDelete
  4. I am a proud owner of a dual 1.42Mhz G4 MDD Mac. I love this thing and am sad that it is no longer supported. Such is technology, but newer is not always better. All these years later I am still upset at Motorola and their abandoning of Apple.

    ReplyDelete
  5. I'm still rockin' a powerbook G4 1.5Ghz and it runs everything I need it too and it's 2016 now! Compared to to my 2006 Intel imac, this older G4 laptop is sadly, faster!

    ReplyDelete