Multi-Core? It’s No Panacea

Robert Colwell, a former chief microprocessor architect at Intel, says
much-hyped dual- and multi-core chips aren’t the final or even best solution
to improved system performance.

“If there is a crying need in the general population for greater
performance, and you become aware of that and you fill that need, you should
reap the rewards and make all the money,” Colwell told
internetnews.com.

But Colwell said the impetus to move to dual- and
multi-core is far more driven by the inability of chip companies such as Intel
and AMD to keep ever faster processors cool enough to run safely and efficiently.

“If I wanted to speed up my computers, the first thing I’d do is add
memory, then get a better chipset, then better 3D graphics, and a faster
hard drive,” said Colwell. “Intel is pumping up the processor because that’s
what it owns. I would really like, for example, attachments I get in my
e-mail to open faster, but that’s not a function of the CPU.

“Other parts of
the system will improve, but they aren’t controlled by Intel, and the
software to take advantage of dual-core isn’t really there yet.”

Colwell, who last worked at Intel in 2001, authored a recently published
book on his experiences there called “The Pentium Chronicles: The People,
Passion, and Politics Behind Intel’s Landmark Chips.”

In an interview he addressed several current issues related to his former
employer. He characterized AMD’s more recent competitiveness good for Intel
and the industry.

“It’s called competition and capitalism,” said Colwell. “From 1995 to
2000, Intel strode the world like a behemoth; we didn’t really have any
competition and that’s when you get stupid and lazy. You get used to the
idea anything you ship will and should be accepted because there’s no
alternative.”

His guess is if AMD keeps executing it has a good chance at achieving 25
percent market share in the next few years.

“They have some great designers,
there, I think they will do well,” said Colwell. “For Intel to maintain 85
to 90 percent share is an unnatural position. But I don’t think it will ever
be 50/50 because Intel has too much expertise in fabs and manufacturing.”

For the Itanium, Intel’s high end processor it started designing over a
decade ago with HP, Colwell sees trouble ahead.

“If you don’t ship in volume
in the silicon business, you will lose to whoever does. That was the lesson
of Alpha, which was a damn good architecture, better than anything Intel had,
but what the world cares about is running software.”

Alpha, developed by the long-defunct Digital Equipment Corp., was a
64-bit RISC microprocessor released in the early 1990s. Alpha
ultimately failed, though it had strong supporters and by some estimates was a
multibillion dollar business.

Colwell said his main motivation for writing the book is that when you
design something as complicated as a microprocessor you don’t really get to
do it again — either it’s successful or it’s not.

“I want to give readers some
insight into the process and hopefully there are some lessons there to be
learned from what happened.”