Chip rivals joust over speed ratings

Intel has funded a new report from Aberdeen, "AMD's Gigahertz Equivalency: Inexperienced Buyers Accept Bad Science," which heavily criticizes AMD's model-number system as confusing to consumers and as "not justifiable in the benchmark science." The report says that "AMD...must soon retreat from the gigahertz equivalency positioning and take another performance-rating approach."

AMD called the report a "shady marketing deal" that distorts the facts.

"They get loads of things wrong in there," said an AMD representative, citing the Aberdeen report's assertion that the Athlon XP model numbers are designed to correspond to Pentium 4 clock speeds. "We've always very firmly said that the numbers compare with the previous generation of Athlon, for consistency," the representative said. "We've worked hard at explaining it and making it clear."

The spokesman also said that Aberdeen did not contact AMD in the course of making the report. "If you're writing about somebody, you'd tend to ask their opinion," he said.

AMD says officially that its model numbers reflect differences in performance between its current Athlon XP processors, based on the "Palomino" core, and the earlier Athlon chips. Athlon XP introduced the "Quantispeed" architecture, which AMD argues makes clock speed less relevant because the chip is executing more instructions per clock (IPC). Therefore, an Athlon XP 2100+ running at 1.73GHz, would achieve performance equivalent to a 2.1GHz version of an older Athlon.

However, Aberdeen's report argues that "clearly the competitive comparisons are to Intel's microprocessors," positioning an Athlon XP 2100+, for example, as equal to a 2.1GHz Pentium 4. That comparison will quickly grow more erroneous as benchmarks, the operating systems and applications evolve, Aberdeen argues.

"The key flaw is that the equivalency rating is a snapshot in a moment in time--and time surely marches on in the computer industry," the report says. "There is a distinct 'Pinocchio factor' that will only grow over time as pseudo-equivalency gradually becomes patently inaccurate."

Although Intel emphasizes that it had no influence on the content of the report, the conclusions closely mirror Intel's own position on measuring processor performance.

There is no single industry-standard method of measuring the performance of a processor, since chips with the same clock speed may deliver different performance levels depending on the application. Intel's own 1.7GHz mobile Pentium 4, for example, delivers roughly the same performance for productivity applications as a 1.2GHz mobile Pentium III.

However, Intel argues that at least clock speed is an objective measurement that cannot be disputed. "Gigahertz is a fact, it's not going to change," said Matt Dunford, Intel performance benchmarking marketing manager. "However, it isn't a faultless measurement of performance."

AMD's insistence on instructions per clock is also misguided, Dunford says, since the Pentium chip has to execute fewer instructions than the Athlon XP. "IPC is no better than gigahertz," Dunford said.

Both Intel and AMD subscribe to the Standard Performance Evaluation Corporation's SPEC CPU2000 benchmark, but Dunford said it would be "misleading" to use its figures as a replacement for clock speed.

Dunford also criticized AMD's "True Performance Initiative," announced last year, noting that the SPEC standard already has wide industry support. "If you're talking about an industry consortium to measure performance, (SPEC) looks like an industry consortium," he said.

Matthew Broersma reported from London.