Pushing the Limit


When it comes to computers, faster doesn't always mean better.

For years, processor speed was the magic number that dictated how well a machine performed. But computers have evolved into much more complicated beasts, and measuring performance no longer boils down to a single number.

Unfortunately, the computer world forgot to tell the rest of us. Advertisements still scream speed, touting the newest, fastest processors.

"The speed of a processor is no longer directly related to its performance," said Renato Figueiredo, assistant professor of electrical and computer engineering at Northwestern University. "The ultimate measure of a computer's performance is how fast it runs software ... and speed is just one piece of the puzzle."

The microprocessor is the heart of a computer, crunching numbers and directing electronic traffic. A microprocessor's speed is measured in hertz, which is the number of cycles it can perform every second.

A 1-megahertz processor performs 1 million cycles per second, and today's speediest processors reach as much as 2 gigahertz, or 2,000 megahertz.

Today's chips run really, really fast.

Because processor speed has increased much more quickly than the speed of other components such as random access memory, or RAM, which stores frequently used data for easy access, a processor often must wait for information to reach it. That hurts performance.

Also, how much work a processor can actually do each cycle is as important as how often it performs that cycle.

Older chips performed a single operation per cycle, so the relationship between raw speed and actual processing power was relatively close.

Newer chips such as the Pentium 4, AthlonXP and G4 perform a varying amount of work per cycle. How much data the chip processes depends on the design of the chip and the software being run on it.

That means a slower chip doing more work per cycle or retrieving data from RAM faster could outperform a faster chip if the faster chip was doing less work each cycle or snatching data more slowly.

For companies such as AMD and Apple, whose chips are designed differently from and run at slower clock speeds than Intel's zippy Pentium 4, convincing consumers that their chips make up for slower speed with other features is a big issue.

Both companies have made various attempts over the years to debunk this so-called megahertz myth.

With the launch of the AthlonXP, AMD wants to persuade consumers to evaluate chips by a "model number" instead of megahertz.

AMD said the model numbers of its new chips show the processor's "relative performance," or how the chip stacks up against a competing processor running at the model-number speed.

For example, the AthlonXP 1800+, AMD's top-of-the-line model, actually runs at 1.53 GHz and is supposed to outperform competing chips running at 1.8 GHz, according to AMD.

"There needs to be new metric for processors," said Patrick Moorhead, vice president for customer advocacy at AMD. "We'd like to see a system that would allow end users to make a more sophisticated choice."

Though AMD enjoyed beating Intel to the 1-GHz mark, currently its fastest processor is about 500 MHz slower than the 2-GHz Pentium 4. Now AMD wants to downplay megahertz.

"We're not trying to hide anything," Moorhead explained, "but we aren't going to lead with a proxy for performance that doesn't mean as much as it used to."

Intel, which produces the high-speed Pentium 4 and enjoys about a 77% market share, won't talk about AMD's recent efforts. But Intel spokesman Robert Manetta agreed that megahertz "is just one of the ways you can evaluate a processor's performance. We support clarifying to the consumer

Apple, which has been trying to downplay megahertz for years, is now "trying to move beyond stats of all sorts," said Greg Joswiak, senior director of hardware product marketing. "It became amazingly not meaningful to compare [processor speeds] across platforms."

Joswiak said Apple is concentrating on providing software to users "who want to exploit a digital lifestyle" instead of focusing on hardware. "People don't want to buy a stat; they want to get something done," he said.

But sometimes it's hard to tell what kind of system a person needs.

"How can you market something that has four or five different features that need to be compared?" Figueiredo asked. "There are so many components that don't get factored in ... it's hard for the average user to understand."

For multimedia tasks such as streaming audio, streaming video or video editing, having fast RAM--and enough of it--is important to getting good results, Figueiredo said. "If your program is bringing a lot of data from memory, it doesn't matter how fast the processor runs each operation; it matters how fast the data can come."

Gamers would benefit from a powerful processor and good video card and sound card. Internet users need a faster connection or modem more than they need a faster computer.

What a user plans to do with a computer is key to determining what kind of components to buy.

Still, there's no tried-and-true way to measure performance, especially across the wide variety of applications a computer could be used for. Intel, AMD and Apple all agree that independent benchmarks that test a system's performance with a variety of applications are the best way for a consumer to decide what kind of system is needed.

But don't expect to find a suite of benchmarks comparing computer systems in advertisements or on store shelves any time soon.

Joswiak and others don't "see megahertz going away. We just need to find a way to meaningfully compare performance across platforms."

Copyright © 2019, Los Angeles Times
EDITION: California | U.S. & World