Aug 23, 2006

Unseen PowerPC: The Cores That Didn't Make It

Ever since Apple, IBM, and Motorola jumped into bed together twelve years ago, tech media—and Apple watchdogs especially—have had a field day with speculation, rumors, and actual news regarding new PowerPC projects. Apple gave a familiar face and a flair of iconoclast to the affair while IBM lent a grave sobriety. This thing could really happen, then, someone to challenge the Microsoft/Intel duopoly. And the stories just kept coming, well into the next decade. But not all of them were so real.

So what were these projects that came and went faster than a Quad Xeon Mac? Handily enough, they're right below, broken down by project. Read on to find out what Apple, IBM, and Motorola had in store for us throughout the Nineties and what didn't make the cut. Through all the rumors, one thing was certain: AIM never had a lack of imagination, even if it didn't always end up in silicon.

Too Much Too Soon

IBM and Motorola jointly announced the 64-bit PowerPC 620 at the October '94 Microprocessor Forum nine years earlier than the Power Mac G5 and its PowerPC 970 processor. Designed for servers, the 620 supported up to 128 MB L2 cache(!) and was set to scale from 133 to 150 MHz. It eventually shipped, albeit briefly, in systems from Groupe Bull. Structurally it was nearly identical with the PowerPC 604, save for some nips and tucks and wider registers and data paths. Think of a PowerPC 604 with twice the lung capacity.

The 620's similarity to the PowerPC 604 was also its undoing. At this point in time, the 604 was clocking up, and because it was a 32-bit chip, it ran cooler and cheaper than its 64-bit brother and actually out-performed it. The PowerPC 604 was also getting upgraded to the 604e, which further blew the pants off the 620. That, plus the fact that no one really needed a 64-bit workstation at that point in time resulted in one dead-end PowerPC core. The chip supposedly reached 200 MHz with an enhanced core (the PowerPC 620e?) before the guillotine fell.

Oops, looks like we didn't need 64-bit that badly yet.

He Goes Both Ways

Talk about ambitious. In '96 sources started to talk the PowerPC 615 that ran x86 instructions natively. How it did this wasn't exactly clear: At first the chip had to boot in either PowerPC or x86 mode, then it gained the ability to run the other ISA's instructions after a five clock-cycle flush, then it decoded the instructions on the fly, and finally it was supposed to have a separate unit or co-processor that made them magic happen. On top of that it plugged into existing OverDrive sockets but also created enormous amounts of heat.

Whew, huh?

That's what Microsoft said when talking to IBM about it. The chip would need Windows NT to succeed and MS didn't think it was a good idea. With nothing else really wanting to run on it—because why would you run x86 software on a chip that didn't do x86 as fast or as cheap as Intel, and who's going to use the PowerPC ISA?—IBM found itself without an audience on this one. Big Blue actually taped out some test units, but they're probably collecting dust or jingling against a set of keys now. This hybrid beast wowed its engineers and no one else.

The So-Called Speed Demon

Exponential, Inc. took the PowerPC 604 design and stripped it bare in order to blast its clock speed into the stratosphere. Using BiCMOS technology that allowed for faster gate switching, the PowerPC X⁷⁰⁴ was slated to run at speeds of 533 MHz in late 1996 when AIM and Intel were toying with 200 MHz. Of course, Mac users ate it up: PowerPC was at least holding even with Intel's offerings and they were eager for more good news. Apple had even invested in the company, and–

And then everyone woke up. The Mach V and PowerPC G3 were already hitting the marks claimed by Exponential, who had in the meantime only delivered a 410 MHz part that didn't so much run as it did that thing where you cross the street in front of a car and try to act like you're running. Yes, BiCMOS pushed the clock up but at the cost of actual performance, power usage, and price—something we would see years later from Intel. The chip failed to run beyond 410 MHz and Apple decided not to use it. Exponential fell apart within the next couple years.

Whither the Motorola G5?

Like many of Motorola's other projects, what would have been the PowerPC G5 ended up a sputtering mess, leaking brain-share and focus. After the failed or postponed revisions to the G4, supposedly the "Apollo" 7460 series, things got icky between Apple and Motorola. The existing G4, the PowerpC 7400 series, was stalling out on its way up the clock ladder and Apple was fuming. Motorola simply couldn't be budged, though. The embedded space didn't need faster chips three or four times a year, so Apple had to deal.

After the messy transition from Motorola's spin-off of its semiconductor business into Freescale, the G5 probably became one of several products. Just which project, however, depends on whom you ask. It might have become the PowerQUICC III, the 8600, or even the as-yet unreleased PowerPC 8700. There's really no telling. In fact, beyond the bevy of cache, clock, and bus upgrades no one really knew what to expect of the G5. Multiple cores, HyperTransport,

Big Blue's G4

After Motorola failed to push the G4 to 1 GHz in a timely manner and repeated themselves on the way to 1.5 GHz, rumors whirled of a revved-up G3 from IBM that would pick up where Motorola had left off. Not only was this revision G3 looking at a Velocity Engine, it would have also gained a faster bus, longer pipeline, better power management, and enhanced multiprocessing as well. But around the time the VX was scheduled to debut and IBM was failing to push the clock on the G5, plans seem to have changed.

Instead of the VX, IBM ended up releasing the much more conservative 750GX, which doubled the on-chip L2 cache to 1 MB, tweaked the cache and bus, and pushed the chip clock to 1.1 GHz and the bus clock to 200 MHz. By the time IBM's PowerPC 750GX arrived in the same timeframe that the VX was expected in, it was clear that making Apple dependent on IBM, who was well into pulling their own Motorola with G5 updates, was not a wise decision and Apple stayed with the G4 for their low end until they could move to Intel.

Too bad Apple hadn't pushed for this during the 500 MHz Fiasco.

Huffing Vapor

Due about a year ago, the PowerPC 350—and its purported follow-on, the 360—would have done everything except wax your back, all in a nice cool little package for portables. Rumors about the 350 said it was supposed derive from the Cell processor and run separate units of the processor across a high-speed bus that could handle up to 128 GB/s. If you think someone was smoking some serious drugs, congratulations, you're smarter than the Mac rumor sites. The 350's hype was a lot more substantial than anything IBM ever spent time working on.

What ended up materializing was... Well, nothing. There never was a portable-optimized PowerPC from IBM, nor one that was a completely new micro-architecture, nor anything on a 65nm process. In fact, it's pretty safe to say that Apple would have to have been selling five times the volume of Macs to make IBM sit up and make a new ultra high-tech core for them like this one. And since the console makers did move that many units, and we have IBM in bed with them now, it's pretty clear where R&D was looking. It wasn't at Apple or a PowerPC 350.

The PowerPC 350 was vapor, pure and simple, no matter how much the rumor sites were huffing it.

980 Ways To Keep Fooling Yourself

IBM was contracted to create four generations of microprocessors for Apple over the period of five years. Yeah right. If you believe that, I have a sentient Power Mac G6 to sell you. In the wake Steve and IBM breaking their promise to deliver a 3 GHz, rumors of the 980 began swirling. There were two main thoughts on this core: The 980 would either be an evolution of the 970 itself, or it would be a stripped-down version of the Power5 as the 970 had been of the Power4. It was supposed to debut at 2.8 to 3.2 GHz and top out at 5 GHz as well.

We'll never know, because there aren't enough online farmhands to sort through all the horse nuggets.

What is obvious is that not long after the 970 debuted IBM stopped caring about Apple's needs on the desktop. If there ever was a contract for future PowerPC cores, it ended after IBM called Steve and said We don't think 3 GHz is gonna work out. So as engineers were taken off the 970 project and put back on the Power5 bus, or moved to the Xenon or Cell teams, the 980 and its successor, the 990, became more and more imaginary until it was completely in the domain of fiction writers and Mac OS Rumors.

The End of the (PowerPC) Line

Well, here we are at the end. The end of both the rumors about PowerPC cores for Apple and the end of the PowerPC era there altogether. As one could easily notice, the failed PowerPC projects became more nebulous and shadowy as they progressed: Someone out there probably still has a PowerPC 620 system running in their closet, and Gil Amelio really did sit down at a 410 MHz X⁷⁰⁴-based Power Mac for an afternoon. But the PowerPC 350 and 980 were nothing more than the fetish of an AppleInsider source. So why the change from sampling to story?

At first, when PowerPC on the desktop was something a lot of people in high places believed in, projects abounded. It was the Cambrian Explosion of silicon. A lot of designs came and went in the same time other designs got a clock-bump. As time went on, however, the desktop gave way to embedded space, variations ceased, and innovation stagnated. Almost everything else from Freescale these days, for example, is descended from the 603. IBM has taken a more aggressive approach but it took Microsoft and Sony to get them to kick it into high gear.

Now, with trumpeting the Power Architecture from the rooftops, one might wonder if a reversal is in order. PowerPC is certainly on its way to replacing MIPS as the ubiquitous load-store ISA that runs everything from your car's throttle to your favorite gaming rig to the mainframe processing all of your sales. But none of those are a return to the desktop. Try telling that to the Mac faithful, from whose breast hope springs eternal. The foggy PowerPC rumors were nothing more than wishful thinking for a bright future.

Now that Apple's with Intel, the future really is bright. So grab your shades and come along for the ride.

No comments:

Post a Comment