Vol 15, Issue 48

November 26, 2001

Marketing PC Performance

By Kevin Krewell

When AMD first briefed me on its model-numbering scheme, my initial reaction was negative. I was working at AMD when AMD, Cyrix, and NexGen (before it merged with AMD) concocted the performance rating (P-rating), the previous attempt to break away from frequency as a measure of performance. The system used only one benchmark—the Ziff-Davis Winstone96—as the basis for comparing challengers to Intel's Pentium.

AMD used P ratings for the AMD-K5 processor, selling a 117MHz K5 processor as the "AMD-K5-PR133," designed to compete with a 133MHz Intel Pentium processor. The K5 was flawed, however, and soon after launching the K6, AMD dropped the P-rating system. The system eventually fell into disrepute, owing to problems that included biased system configurations, reliance on only one benchmark, products with weak floating-point units, and one vendor who even retargeted the comparison to AMD's K6 instead of to Intel's Pentium.

I still have mixed feelings about any attempt to replace a system based on a scientific measurement with one based on a marketing conceit. Science is supposed to be exacting, testable, and repeatable—and generally accepted by professionals. The concept of megahertz is scientific. One can measure megahertz with appropriate equipment.

"Performance" is a relative term; it must be compared with something else—in some equivalent timeframe. Model numbers are the invention of marketing departments, and because I am often involved with the marketing aspects of technology, I know their abuses. I was particularly concerned because with the model numbers came something called "QuantiSpeed." QuantiSpeed was pure marchitecture—a term coined to characterize the unholy marriage of marketing and computer architecture.

Before you start thinking that I'm merely picking on AMD, let me digress to discuss Intel and the current trend of technology companies trying to mimic consumer companies. An example: combine superpipelining, a microarchitecture design technique, with "hyper-," a marketing euphemism for faster, and you get "Hyper Pipelining." This is also classic marchitecture. Marketing people are smart, but they often struggle to communicate technological advances to the masses. Somehow, simple constructions like "superpipelining" and "very deep pipelining" didn't satisfy Intel Marketing's need for a label. Intel also attached the ambiguous label "NetBurst" to the Pentium 4 processor architecture, because, I guess, Hyper-Pipelining was still too technically descriptive.

The problem for technologists is that efforts like Hyper Pipelining and NetBurst often look like hokum, a con artist's attempt to deflect attention away from real features and performance. These efforts have a tawdry quality, as if the electronics industry were selling technology as Madison Avenue would sell laundry detergent: See the New and Improved Pentium 4 processor—now with NetBurst formula!

AMD didn't start the marchitecture cold war—but it didn't want to lose it, either. So AMD rallied its troops to close the marchitecture gap—along with the clock-frequency gap. Thus, AMD's Athlon XP processor came with the QuantiSpeed secret sauce. One would have to assume that without the QuantiSpeed magic, Athlon would be unable to support the model numbers and would have to rely on antiquated megahertz clock speed for comparisons with Intel.

All this marketing voodoo, on both sides, shows a lack of respect for the consumer's intelligence. Couldn't AMD find a better way to communicate performance without model numbers? Do NetBurst, Hyper-Pipelining, and QuantiSpeed mean anything to consumers? In AMD's case, these terms may even turn off customers that have given AMD the most loyalty—savvy VARs and computer enthusiasts.

Since their introduction, AMD's new model numbers have received little more than lip service; already, we are receiving reports that some small vendors are mistakenly advertising Athlon model numbers as processor frequencies.

Large OEMs have used these numbers inconsistently. MicronPC is following the model-number program strictly (with no listed megahertz metric), but Compaq prominently lists the frequency next to the model number. HP is using the frequency in its build-to-order configuration menus without bothering to list the model number in the processor pull-down menu selection.

Intel will face this performance vs. frequency dilemma with its Itanium processor, with Hyperthreading (there's that hyper thing again; hyper-ness must be in at Intel) technology, and the future Banias processor, although for these products and technologies, performance cannot be measured in megahertz alone. For the moment, Intel needs frequency, because the NetBurst Pentium 4 design reduced instruction-per-cycle performance to less than that of the Pentium III. That was the price Intel paid to ship obscenely fast processors. Eventually, however, Intel will also need a better measurement, but it's unlikely Intel will accept AMD's definition, no matter how reasonable it may be.

For the consumer, I suggest that nothing replaces independent research. Read PC-related magazines such as PC Mag, PC World, and MaximumPC. Visit Web sites CNET and ExtremeTech. Shop around. Don't trust retail salespeople. No single magic performance number will answer the question of which is the most powerful processor.

All the processors available in mainstream PCs today can meet the needs of most buyers. Both the Athlon XP and a 1.5GHz or faster Pentium 4 are exceptional processors. The final choice will come down to consumer confidence and brand preference. Instead of spending energy on hyping model numbers, AMD would do well to spend more time on the rest of the marketing message.