Tech
Tony Smith: Macs marred by Megahertz
If you want to make it big, clock speed is what you need
Back in January, at the Macworld Expo in San Francisco, Apple iCEO Steve Jobs happily told the keynote crowd how the colour of a computer was way more important to consumers than the machine's megahertz rating.
Oh, how we all laughed in agreement. Oh, how Jobs' words are coming back to haunt him -- and us -- less than a year later.
Why did we nod in agreement with Apple back then? Because, in addition to launching the iMacs, Jobs had already debuted what he called "the world's fastest PC", the blue-and-white Power Mac G3, with its -- gasp -- 400MHz PowerPC 750 (G3) processor. At the time, the best Intel could field was a 500MHz Pentium III, and the Mac faithful knew the 400MHz G3 -- the chip, not the computer -- was just as fast, if not faster.
How times have changed. Intel's arch enemy, AMD (what, you really think Apple's bothered by Motorola?), has just shipped the 700MHz version of its Athlon processor, and Intel itself is readying the release of a 700MHz Pentium III. And what does Apple have?
A brand spanking new Power Mac G4 may be a really cool machine, but it has to compete with a 450MHz PowerPC 7400 (G4) processor. For all the earlier "world's fastest PC" and "first desktop supercomputer" hype, Apple's computers are suddenly looking rather underpowered.
When 450MHz isn't 450MHz
Now, the PowerPC diehards will be quick to point out that the 7400 may only clock at 450MHz - we'll ignore the 500MHz version for now, since Motorola hasn't been able to get the thing to work properly - but it still offers comparable performance to the Pentium III.Well, sort of. A variety of third-party Web sites have done the numbers and found that for most applications the 400MHz G4 offers the same performance as a 500MHz G3. And since the latter offers comparable performance to a 600MHz Pentium III, the 450MHz G4 won't be much slower than the 700MHz Intel chip. And the 500MHz G4, when it actually gets to that speed without corrupting the contents of its data cache, will be much the same. From a technical standpoint that's a sound argument. However, from a marketing perspective it plain sucks. Apple's real problem on the processor front is not that it isn't Wintel based -- the success of the iMac with Wintel and first-time buyers proves that -- but that the PowerPC line isn't perceived as being as fast as the Pentium. Why? Because, despite Jobs' comments, PC buyers do consider megahertz the chief yardstick by which a computer's performance can be measured. Even Motorola seems to have finally figured this one out -- although not without some prompting from Apple, I suspect. At last week's chip industry shindig, the Microprocessor Forum, held in San Jose, California, Motorola revealed it is developing a second-generation G4 chip to ensure the line is better able to compete on clock speed. It's clear that the decision to produce the G4-II was made only recently -- evidenced by Motorola's September PowerPC road map, which lists the G5 (the multi-core G4-based chip codenamed V'ger) as the next release. The G4-II increases the number of stages in the pipeline through which an instruction must pass to be processed to accommodate the 700MHz (that number again) and higher clock speeds that Motorola plans to support. Adding the extra stages will, to a degree, reduce the speed advantage the PowerPC has over the Pentium III at equivalent clock speeds, but it will allow it to go back to offering chips within 50MHz of Intel's top frequency. Of course, Motorola gave no indication when the G4-II will appear, beyond saying it will spill the shipment beans in early 2000. Optimistically, the company could have 700MHz-and-up G4s out by this time next year -- given that the G4-II is a completely new microarchitecture and not a G4 with extras bolted on - by which time Athlon and Pentium could easily be up to 900MHz, possibly even 1GHz. Meanwhile, of course, IBM is continuing to develop its own G3 line and there's the very real possibility that an IBM G3 with the latest silicon-on-insulator technology (along with copper interconnects) that it (and Motorola, for that matter) is already using, could be faster than even the 500MHz G4. True, support for the G4's AltiVec system (or Velocity Engine) will make a difference, but not much since its usability with most applications is limited. And since the customized G3 chip IBM is developing for Nintendo's next-generation console is likely to support AltiVec -- it will have to, since Sony's Emotion Engine processor, the heart of the PlayStation 2, has a vector-processing engine of its own -- even that advantage may ultimately be limited. Safe in the water
For the time being then, Apple has a problem. Until Motorola ships the G4-II, it's going to have a tough time coming up with machines that run much above 500MHz. It can offer a 600MHz G4, but for all the extra 100MHz it won't be significantly faster than the 500MHz machine. Or Apple can attempt to persuade people that multiprocessing on the Mac really works by releasing a multi-CPU machine and hoping they don't notice that the speed advantage really isn't that great -- even with multiprocessing Mac OS X, doubling the number of CPUs won't double performance. And until Motorola gets the current G4's bugs fixed, Apple isn't even going to be able to offer a 500MHz Mac that actually runs at that frequency. And until Motorola ... But that's the point: Apple is entirely dependent on one company, which is what the PowerPC Alliance was intended to prevent. With IBM out of the desktop market, Apple has to rely on Motorola to deliver the chips it needs, when it needs them and at clock speeds that it can realistically take to market. And Apple does need chips it can use to compete with Wintel. Indeed, now that the Mac OS has largely lost its superiority over Windows it needs it more than ever. Sorry, Steve, but higher clock speeds, not cute colour cases, are what's needed now.
Now, the PowerPC diehards will be quick to point out that the 7400 may only clock at 450MHz - we'll ignore the 500MHz version for now, since Motorola hasn't been able to get the thing to work properly - but it still offers comparable performance to the Pentium III.Well, sort of. A variety of third-party Web sites have done the numbers and found that for most applications the 400MHz G4 offers the same performance as a 500MHz G3. And since the latter offers comparable performance to a 600MHz Pentium III, the 450MHz G4 won't be much slower than the 700MHz Intel chip. And the 500MHz G4, when it actually gets to that speed without corrupting the contents of its data cache, will be much the same. From a technical standpoint that's a sound argument. However, from a marketing perspective it plain sucks. Apple's real problem on the processor front is not that it isn't Wintel based -- the success of the iMac with Wintel and first-time buyers proves that -- but that the PowerPC line isn't perceived as being as fast as the Pentium. Why? Because, despite Jobs' comments, PC buyers do consider megahertz the chief yardstick by which a computer's performance can be measured. Even Motorola seems to have finally figured this one out -- although not without some prompting from Apple, I suspect. At last week's chip industry shindig, the Microprocessor Forum, held in San Jose, California, Motorola revealed it is developing a second-generation G4 chip to ensure the line is better able to compete on clock speed. It's clear that the decision to produce the G4-II was made only recently -- evidenced by Motorola's September PowerPC road map, which lists the G5 (the multi-core G4-based chip codenamed V'ger) as the next release. The G4-II increases the number of stages in the pipeline through which an instruction must pass to be processed to accommodate the 700MHz (that number again) and higher clock speeds that Motorola plans to support. Adding the extra stages will, to a degree, reduce the speed advantage the PowerPC has over the Pentium III at equivalent clock speeds, but it will allow it to go back to offering chips within 50MHz of Intel's top frequency. Of course, Motorola gave no indication when the G4-II will appear, beyond saying it will spill the shipment beans in early 2000. Optimistically, the company could have 700MHz-and-up G4s out by this time next year -- given that the G4-II is a completely new microarchitecture and not a G4 with extras bolted on - by which time Athlon and Pentium could easily be up to 900MHz, possibly even 1GHz. Meanwhile, of course, IBM is continuing to develop its own G3 line and there's the very real possibility that an IBM G3 with the latest silicon-on-insulator technology (along with copper interconnects) that it (and Motorola, for that matter) is already using, could be faster than even the 500MHz G4. True, support for the G4's AltiVec system (or Velocity Engine) will make a difference, but not much since its usability with most applications is limited. And since the customized G3 chip IBM is developing for Nintendo's next-generation console is likely to support AltiVec -- it will have to, since Sony's Emotion Engine processor, the heart of the PlayStation 2, has a vector-processing engine of its own -- even that advantage may ultimately be limited. Safe in the water
For the time being then, Apple has a problem. Until Motorola ships the G4-II, it's going to have a tough time coming up with machines that run much above 500MHz. It can offer a 600MHz G4, but for all the extra 100MHz it won't be significantly faster than the 500MHz machine. Or Apple can attempt to persuade people that multiprocessing on the Mac really works by releasing a multi-CPU machine and hoping they don't notice that the speed advantage really isn't that great -- even with multiprocessing Mac OS X, doubling the number of CPUs won't double performance. And until Motorola gets the current G4's bugs fixed, Apple isn't even going to be able to offer a 500MHz Mac that actually runs at that frequency. And until Motorola ... But that's the point: Apple is entirely dependent on one company, which is what the PowerPC Alliance was intended to prevent. With IBM out of the desktop market, Apple has to rely on Motorola to deliver the chips it needs, when it needs them and at clock speeds that it can realistically take to market. And Apple does need chips it can use to compete with Wintel. Indeed, now that the Mac OS has largely lost its superiority over Windows it needs it more than ever. Sorry, Steve, but higher clock speeds, not cute colour cases, are what's needed now.