AMD plans to change its Athlon naming scheme to better compete with Intel's high-speed Pentium 4, according to published reports. Executed carefully, the supposed plan could be a real boon to the megahertz-strapped chip maker, analysts say.
AMD officials decline to comment on the report, which says the company will in October rename its flagship Athlon processor, attaching numeric codes that correspond to similar-performing P4 chips. Intel's P4 chip, now running at up to 2GHz, has far outpaced AMD's 1.4GHz Athlon in terms of frequency.
But most benchmarks, including PC Advisor's, show the Athlon's overall performance is comparable to the higher-frequency P4 unless the test PC is running Windows 2000. Under the scheme, names of new Athlons might reflect that.
Mike Feibus, principal analyst with Mercury Research, couldn't discuss or confirm AMD's plans, but he acknowledges the company's uphill marketing battle with the Athlon. It is a hard sell when your product appears to offer less performance than it really does, he says.
The P4 appears to outperform the Athlon because PC users typically look at frequency as the primary measure of performance, but it is actually only part of the equation, he says. The amount of work you do per clock tick is just as important as how fast the clock ticks, he says. The Athlon does more work per clock tick than the P4, but the P4 runs its clock faster.
Feibus uses a running water analogy. The Athlon collects, then empties, water using a bucket. The P4 uses a cup to collect the water. The bucket takes longer to fill and it isn't emptied very often; the cup fills up quickly and dumps more often. In the end, both sides move about the same amount of water [performance] in the same amount of time, but they do it in different ways.
If AMD begin naming its chips in relation to P4 speeds, it won't be the first time the company has instituted performance-rated titles, Feibus says. AMD and fellow chipmakers Cyrix and IBM have all used the tactic in the past, with varied results.
The numbers game
Cyrix began using performance ratings in 1995 with its 6X86 chip, which was competing with Intel's Pentium processor at the time. That first 6X86 ran at 100MHz, but carried a PR120 rating, meaning its performance was comparable to the 120MHz Pentium, according to Mercury Research records.
AMD first began using a PR rating with later iterations of its K5 processor, which it launched in 1996, according to Feibus. The K5 didn't do well in the market, but that was because the chip itself was a dud — it wasn't an issue with the performance ratings, he says. The first performance-rated K5 was the K5-PR100, which actually ran at 75MHz.
IBM found perhaps the greatest success with performance ratings, Feibus says. The company, which produced processors for Cyrix, also marketed its own version of the 6X86 in 1997, which later became the M2 processor. IBM convinced Compaq to use its performance-rated chips in one of its first sub-$1,000 PCs and the combination was a huge success.
But performance ratings were shortlived for both AMD and IBM. When AMD launched its K6 line, the new chip's frequency and performance compared favourably to Intel's top processor of the time, its Pentium with MMX. So AMD stopped using performance ratings.
And despite its success with Compaq, IBM stopped producing processors in 1998 after National Semiconductor purchased Cyrix and began producing its chips. Under National Semi, Cyrix continued to use the performance ratings.
Perfomance-rated's bad rap
When AMD, IBM and Cyrix all used performance ratings, PC vendors grew to accept them, Feibus says. However, after IBM and AMD discontinued the naming convention, Cyrix started to play fast and loose with its own ratings, causing vendors to distrust them, he adds.
"Their [PR] definitions changed all the time and the company seemed to alter them simply to match their products to Intel's roadmap, he says. "Cyrix ran into production problems and fixed them on paper," he says.
Cyrix, now owned by Via Technologies, shipped its last performance-rated processors last quarter.
While Cyrix's actions may have frustrated PC vendors, customers are not sour on performance ratings, Feibus says. And as IBM's former success shows, they are not averse to buying products with such ratings. So, if a company like AMD decides to resume using ratings, PC buyers would probably be interested — as long as the PC vendors use the renamed chips, he says.
Ideally, some sort of system should be in place to make sure CPU makers wouldn't abuse the system. "There have to be controls there, and it is do-able," Febius says.
In the end, a closely watched system of performance ratings could benefit PC buyers trying to make the best decision when it comes to PC performance, like looking at the badge on the back of car.