Microprocessor vendors have begun moving away from describing their chips with the sort of nerdy “speeds and feeds” metrics that have dominated computing for few decades. It’s part of a dramatic sea change in how PCs, tablets and smartphones are evaluated, bought and sold.
In fact, the notion of computer “performance” is being completely redefined. Instead of chip vendors worrying about tweaking their processor and graphics performance to eke out a few more frames per second on the latest games, they’re worrying about how to make trackpads more responsive and how to make a laptop start up faster after being shut down.
Tablets and phones are equally dependent on microprocessors, but they’ve never really been sold on the basis of chip benchmarks – or even what chips are in them!
A New Look At Power
This week, Intel hosted thousands of developers at its Intel Developer Forum in San Francisco, discussing everything from software-defined radios to touch and voice control to the evolution of the data center.
And – oh yes – there was “Haswell,” Intel’s fourth-generation of what it calls the Core chips for desktop PCs.
Normally, the Haswell presentation would have been packed with roadmap slides, celebrations of clock speeds, a picture of the processor die and the “wafer shot,” where an Intel executive would triumphantly hold aloft the first circular wafer containing the bare processor dice. A year ago, Intel touted a level-3 cache, memory and string performance enhancements, and a dedicated random number generator within its new “Ivy Bridge” chips (the third-generation Core chips, out now).
Forget that. This week, Intel’s technical braggadocio boiled down to offering the same performance as Ivy Bridge, at half the power. There were no clock speeds, no cache sizes or instruction accelerators. Instead, Intel executives positioned Haswell as the foundation, not the focus. “Delivering a processor is not enough; at the end of the day, it’s about the software,” said Dadi Perlmutter, the head of Intel’s chip group.
Redefining Performance
A few blocks away, executives at AMD were saying similar things. There, Leslie Sobon, AMD’s corporate vice-president of marketing and head of product, quickly flipped through a presentation on the company’s APUs (accelerated processing units), which integrate a microprocessor with graphics processing. Instead, she was eager to talk about how much of the world, with the exception of the U.S., Japan, and Germany, is moving away from worrying about whether or not a chip runs at 2.8GHz or 2.9GHz.
“Low power – that’s good performance,” Sobon said. “Good battery life – that’s now considered performance.”
At the same time, also in San Francisco, Apple was launching its iPhone 5. And what do we know of its processor, the A6 – except that it’s better than than the A5 iPhone 4S? Very little, according to Apple, the new A6 promises “up to” two times the CPU and graphics performance; the A6 is 22% smaller than the A5; and the battery life of the new iPhone is slightly better than the battery life of the 4S (how much of that is due to changes in the battery is unclear#). Not a hard spec to be seen.
Designing For Consumers, Not Engineers
A key reason for the change is the attempt to market devices in ways that are meaningful to consumers. When everyone was running productivity software on PCs – there were common benchmarks that did a decent job of explaining how fast a machine would perform common tasks.
Today, there are so many different applications running on so many different platforms, that most benchmarks don’t make much sense anymore. (The possible exception – PCs for gamers, who have very specific needs around graphic peformance.)
So it’s a new world in the computer market, described in the more abstract language of “experiences,” rather than in bar charts and graphs.
“I think that people’s performance expectations are different than what the computer market is selling,” said Mike Feibus, principal analyst at TechKnowledge Strategies. “And they were slapped in the face with this with the iPad in 2010. This is how people perceive performance: if [the system] comes up right away, if things move quickly.”
Basically, the message is this: chip clock speed doesn’t matter. Power usage matters. How fast a computer boots up matters. The responsiveness of a user’s touchpad matters. For the majority of users, just about any computer can run all the software that a user needs at an adequate performance level. That’s “good-enough” computing, and it’s evolved from the desktop to find a home on all kinds of devices.
“That’s not to say raw [processor] performance isn’t important; it still is,” Feibus said, especially with applications like photo editing or voice recognition. But the industry needs new ways to sell the traditional “good, better, best” comparison, he said.
This goes beyond mere marketing. Intel internally believes that the company may have hit a plateau in processing needs, which means that its engineers have no choice but to focus on lowering power and other aspects of the computing experience. The company’s internal market researchers have produced data that shows potential customers will start to issue poorer ratings to a notebook whose trackpad’s latency, or “lag,” goes beyond 250 milliseconds, a source at the company said.
So Intel has rethought its design approach. “For a long time, we worked from the inside out,” the Intel source said. “We developed the best [microprocessor] engine we could and said, here you go. We don’t know what you’re going to do with it, but go party. Now, we recognize that if you don’t work from the outside in – if you’re not thinking about experience from, well, the first moment, you’re setting yourself up to, well, possibly to fail.”
Fortunately for chip makers, the same forces that pushed them to faster and faster clock speeds also help them lower power consumption. That would be the famous Moore’s Law, which technically states that the number of transistors on a given chip doubles about every 18 months. In reality, Moore’s Law gives chip designers a range of choices: improve the chip’s computational performance, lower the power, or some combination of the two.
Burying “Intel Inside”
Intel made that shift a couple of years ago, rebranding its Core line into the Core i3, Core i5 and Core i7 lines. And then there’s the Atom, an X86 chip that can fit inside smartphones, tablets or netbooks. But Intel’s “i” designations still hide a great degree of variation between the individual processors within each family.
Years ago, PC makers trumpeted exactly what chip was powering each PC, even going so far as to disclose details like the memory speed. The race to 1-gigahertz chips was a major event. With today’s ultrabooks and tablets hardware makers seem reluctant to divulge details beyond just the processor family.
Buyers still have to know the difference between an x86 chip and an ARM processor, although the line is blurring with the launch of Windows RT, which runs on the cheaper, lower-power ARM chips, and Windows 8, which uses x86 silicon. But it’s increasingly doubtful that consumers will give a hoot about the differences between a dual-core Core i5 3470-T and a quad-core i5 3330.
Device makers seem perfectly happy about that. Phone and tablet makers instead choose to focus on qualities like screen size, talk time, battery life and operating system. And even PC builders are following suit. Two years ago AMD’s “Fusion” program, Sobon’s brainchild, pulled off the AMD-branded stickers attached to most PCs and let the manufacturer sell the product. This week, at an behind-the-scenes look at one big PC maker’s Windows 8 tablets and ultrabooks, there were no “Intel Inside” stickers to be seen.
The Future Of Chips?
What does this mean for the future? At this point, we just don’t know. PC manufacturers are reluctant to talk about how they plan to price or market their new products before launch.
But one thing seems clear: “Intel Inside” may not go away entirely, but it’s likely to get harder and harder to find out which processor powers the a particular product. A decade ago, PC buyers bought the best collection of parts. Now, the new era of tablets, ultrabooks and convertibles is making individual components – including processors and their specs – increasingly irrelevant.
After all, if chip vendors don’t want to talk in terms of gigahertz, who does?