Beyond MHz and GHz

Selling processors by something other than megahertz speed isn't a new idea. Some companies have used model numbers to mislead the public, while others claim they are helping consumers better understand the real performance of their processors.

There was quite a debate in the 80486/68040 era. Until then, microprocessors ran internally at the same speed they accessed the system bus. With the 486DX design and the '040, the CPU ran at twice bus speed.

Apple stuck with bus speed for quite a while, eventually switching to internal CPU speed when the Wintel world sounded faster. That's why you may see the PowerBook 540c rated as a 33 MHz computer by some source (including LEM) and as a 66 MHz computer by others.

Today we think nothing of rating a CPU by internal MHz speed, not the 66, 100, 133, or 200 MHz but the processor uses to access the motherboard.

From the beginning, we have considered the MHz rating an important indicator of CPU performance. After all, a 40 MHz 68030 ran circles around a 25 MHz 68030. Of course, there were problems with next generation CPUs, such as the 68040 outperforming the 68030 by a factor of roughly 2.5 - in other words, the 20 MHz Centris 610 held its own against the 40 MHz Macintosh IIfx.

Likewise, the G3 turned out to be significantly more powerful than the earlier 604e processor. The G3/233 was roughly equivalent to the 604e/350 - and, as you may remember from the ads, about twice as fast as a Pentium running at the same clock speed.

And that points to the great problem with MHz rating: It doesn't predict performance between different processors.

I recently looked over the various benchmarks for different models, trying to devise a scale that would help quantify the performance difference between the 8 MHz 68000 of the earliest Macs and the G3s and G4s used today. It's not strictly scientific, but here's what I came up with:

• 1: 8 MHz 68000
• 4: 16 MHz 68030
• 6: 25 MHz 68030
• 10: 40 MHz 68030
• 15: 25 MHz 68040
• 40: 60 MHz 601
• 233: 233 MHz G3

By this scale, we see that a 500 MHz iMac or iBook is roughly 500 times more powerful than the original Macintosh, nearly 100 times as powerful as a Mac IIci, and 25 times more powerful than the Power Mac 6100. These are rough figures, but they'll get you into the ballpark.

Of course, that's all using the same brand of computers. It gets messy when we throw the 603/604, G3/G4, Celeron/Pentium, Duron/Athlon, and Intel/AMD comparisons into the mix. Megahertz will tell you which G4 or Athlon is faster than another G4 or Athlon, but it gives no indication whatsoever how either of these compare with the Pentium III or Pentium 4.

What the computing industry needs is some sort of cross-platform performance rating that indicates how well a particular processor performs inside a computer and under an operating system. That said, it should avoid being software specific - you really can't compare Microsoft Office optimized for Windows with the Mac version, let alone make a meaningful comparison between Office on Windows and an Office-equivalent running under Linux.

I can't say just how the benchmark should work, but it should be general enough that it can be easily ported to Windows, Linux, the Mac OS (classic and X), and any other OS people may want to try. It should be processor oriented, so it should avoid tasking the video subsystem, accessing the hard drive, talking to a network, etc. The performance rating should test the CPU, cache, motherboard, and system memory.

It may be that we'll end up with several different benchmarks, just as the SPEC benchmarks floating point and integer performance. One test should emulate number crunching in a spreadsheet, another might work with a large memory-based database, another might simulate word processing, and yet another might specifically look at how well the system rips MP3s or video.

Whether all that will help you get a better price is hard to say, but at least prospective buyers will know exactly what you have without having to await feedback from the seller.

Such a set of performance ratings wouldn't just level the playing field in Mac vs. Wintel comparisons. It would help Mac users quantify the difference between the various G3 and G4 processors; help Intel differentiate the Celeron and Pentium lines; let performance junkies better compare Athlon, Pentium III, and Pentium 4; and possibly answer some of the questions of Mac OS 9 vs. 10 and Linux vs. Windows performance.