Mac Musings

Moore's Law, New Macs, MHz Myths

Daniel Knight - 2001.07.20

Everyone knows that Moore's Law states computers will double in speed every eighteen months. Unfortunately, that isn't exactly what Moore's Law states.

Moore's Law is based on an observation by Mr. Moore, the founder of Intel, that the number of components in a chip doubled every two years. That was later revised to the 18 month figure. By extrapolation, it also predicts that computing power will double on the same schedule.

Moore's Law does have impressive predictive value, even in areas it didn't specifically address. For instance, hard drives seem to double in capacity at roughly the same rate, although Moore's Law has nothing to do with hard drives. Chip density does keep improving, and today's leading CPUs are created on a 0.18 micron scale with a lot of components. RAM chips do double in capacity almost like clockwork, and so does CPU power.

Notice that I didn't say speed, although that's generally true as well. Two years ago this August, Motorola did all they could to produce a 450 MHz G4 chip; this coming August we'll see an 867 MHz version of that chip, and it may offer more than twice the power of the G4/450. That's because Motorola has tweaked the design and now includes an on-chip level 2 cache, giving it a lot more components than the earlier G4 design, along with a 2 MB level 3 cache to further boost performance.

Thanks to AltiVec, the G4 is also able to offer stunning performance for a limited range of tasks - up to twice the performance of a Pentium 4, which runs at double the clock speed, as Steve Jobs demonstrated on Wednesday. That's because power and performance are about a lot more than MHz.

For instance, the 68040 offered roughly 2.5x the performance of the 68030 at the same MHz rating, and the G3 provided somewhere between 1.5 and 2x the power of the 604 processor at the same clock speed. Thus it should surprise nobody that one type of processor can outperform a different one - even if the second has a much higher MHz rating.

Still, we expect the G3 and G4 to advance in MHz rating in accordance with Moore's Law. Double the performance in 18 months translates to a 25-30% increase in MHz speed every six months, the time between Macworld Expos.

How are we doing?

The last Power Mac G3 was introduced 19 months after the first and ran at 193% the original's clock speed. That's on the money.

Over three years, the iMac has gone from 233 MHz to 700 MHz. When the fastest model ships in August, it will offer 3x the MHz rating of the original - and the faster on-chip level 2 cache may further boost performance. Moore's Law predicted 4x the performance after three years, but we're not too far from that mark.

You know, 867 MHz sounds a lot more than 18% faster than 733 MHz. Over two years, Motorola hasn't quite managed to double the MHz rating of the G4. However, the new G4 is more powerful than the older one because of an on-chip level 2 cache, an extra math unit, and some other tweaks. Still, we should have been at this level six months ago according to Moore's Law.

If we look at the Windows side of the world, we see that Moore's Law hasn't failed. The Pentium III has gone past the 1 GHz mark, the AMD Athlon is somewhere in the 1.3 GHz area, and Intel's flagship Pentium 4 has been announced at 1.8 GHz. It almost seems like the competition between Intel and AMD has passed the speed limit implied by Moore's Law.

So what's up with the Mac? More specifically, why isn't the G4 running at 1 GHz and beyond?

That's a darn good question. IBM, Intel, AMD, and Motorola are all using 0.15-0.18 micron technology, so it seems odd that the G3 hasn't broken 700 MHz and the G4 is still shy of 900 MHz. I'm no chip engineer, but if they could come closer to offering MHz parity, that would be one less reason Windows users had to despise the Mac.

The G4/867 outperforms the Pentium 4/1.7 GHz at certain tasks, as Steve Jobs demonstrated at the keynote. It's up to twice as fast - now imagine if a G4/1.7 GHz existed and could offer 3-4x the performance of a Pentium 4 at the same clock speed!

The G4 has a different design philosophy than the Pentium 4. The G4 is meant to handle simple instructions quickly with a 7-stage pipeline. The Pentium 4 is meant to handle more complex instructions and spends more time dealing with them in a 20-stage pipeline. The longer pipeline gives the P4 a lot of power, but it also creates a huge 20-cycle penalty every time the CPU has to clear the pipeline. This is part of the reason the G4 can outperform the P4 despite running at half the clock speed.

Unfortunately, that's not an easy concept to explain when everyone looks at MHz. Higher MHz numbers are better than low ones - that's why the 700 MHz iMac is more desirable and more expensive than 500 and 600 MHz models. But MHz doesn't translate between processors. If they all existed, a 500 MHz 68040, PowerPC 601, G4, 80486, Pentium III, and Pentium 4 would not offer the same performance, because some designs are simply more efficient than others.

For the most part, unless you're working with digital video or huge Photoshop images, computers have been fast enough for several years now. Faster is better and nicer, but for most users it's less necessary than it used to be.

However, when you are up against those demanding tasks, it's nice to know that the G4 can outperform the Pentium 4 despite the apparent MHz handicap.