Some laws are legislated, like the federal No Child Left Behind act. Other laws have no room for changes or amendments, such as the law of gravity. Still others are merely postulates, which apply for a time but then give way to more finely honed hypotheses.
Such may be the case with Moore’s Law
Coined by Intel co-founder Gordon Moore in 1965, Moore’s Law has pretty accurately described every technology transition we’ve had to date. Consider that we’ve gone from the 386, released in 1986 with 275,000 transistors, to today’s Pentium 4 Celeron, which packs upwards of 42 million transistors.
And that’s just regarding common desktop CPUs. Intel has made versions of its 64-bit Itanium with 500 million transistors. (The die sizes of today’s chips are larger than eighteen years ago, so this isn’t a normalized comparison, but you get the idea.)
But now, as ever faster and ever more power-hungry mobile processors begin to displacing pokier desktops, at least one person in the know sees a big brick wall looming along the seeming endless road of Moore’s Law.
“Moore’s Law used to be in some sense free, but no longer,” said Bob Colwell, a private consultant who was formerly Intel’s chief architect for x86 microprocessors. “There’s reason to think that we’re pushing power, performance, and clock rates too hard.”
Colwell vented his contrarian position in a February colloqium before an advanced class of electrical engineering students at Stanford University. (An hour-long video of the presentation can be accessed here. )
There are several reasons Colwell thinks repeal of Moore’s Law is in order.
The most basic will be recognized by anyone who’s ever lamented a laptop’s tendency to double as a space heater. Today’s chips run too hot. The reason is power density. Today’s microprocessors dissipate too many Watts per every square cm of silicon (cm with a superscript of 2).
And since computer cooling fans are already screaming as hard as Harrier jets, there’s not much headroom for building faster CPUs that won’t be at risk from frying.
Looking beyond the power problems that higher clock speeds engender,
Colwell questions the whole rationale for ever-faster microprocessors. “I
think usable CPU performance is no longer following the curve,” he said.
It’s often hard to see the difference between a 300 MHz Pentium II and a
1.0 GHz Pentium 4 on Word and email applications, Colwell pointed out. One
exception is games, but that’s essentially a niche market, he noted. (But it’s one huge niche market though.)
Nevertheless, the industry seems fixated on clock speed, much as
automakers once one-upped each other on horsepower. “This whole game of
pushing clock speeds higher and higher has been so lucrative we’ve been
afraid as an industry to try anything else,” Colwell said.
But in a few years, perhaps as a result of power-density problems–the
clock-speed race is going to end.
In case you think this is a lamentable state of affairs, Colwell has an
alternate viewpoint that should appeal to IT managers feeling hamstrung by
today’s fast–but aging–architectures.
“I want to start adding real value to microprocessors, say, in terms of
wireless or networking,” he said. “If every machine’s going to be networked,
then why not stick the networking paraphernalia on the CPU? Let’s at least
see if that makes sense.”
Although Colwell didn’t come right out and say it, what he alluded to
next in his talk is even more heretical. Namely, Intel’s ubiquitous x86
architecture might not be the best vehicle for transitioning to this brave
new networking-on-the-CPU world.
As Colwell pointed out: “The x86 has so much baggage, you can’t believe
it.”
It’s important to note that he wasn’t casting asperions on the chip’s
designers. (Remember, he was one, once.) Rather, burgeoning complexity is an
inherent part of the x86 architecture, because it has been revised into new
generations for over twenty years. Its designers all the while have had to
jump through electronic hoops to maintain backward compatibility for the
sake of older applications software already in end-users’ hands.
Colwell encapsulated this process: “Intel has succeeded in pushing x86
into higher performance and faster clock [speeds]. The downside is design
fragility The more complicated these things get, the more fragile they
feel.”
He offered an analogy to thoroughbred racehorses, which sometimes break
legs if they run too fast. Similarly, with CPUs, speed comes at the price of
design complexity and inbred bugs.
Whether or not it’s time to repeal Moore’s Law immediately, or in several
years’ time, Colwell’s arguments make sense. Sun Microsystems chairman Scott
McNealy famously coined the phrase, “the network is the computer.” Now,
perhaps it’s time to begin turning the microprocessor into the network
Alexander Wolfe is a senior editor with internetnews.com
