http://economist.com/printedition/displayStory.cfm?Story_ID=3D3321802
http://economist.com/printedition/PrinterFriendly.cfm?Story_ID=3D3321802
Semiconductors
What Intel's latest stumble means for the chip industry's rule of thumb.
I'm so sorry, says Barrett
IT IS not often that the chief executive of one of the world's biggest
companies gets down on one knee and begs for forgiveness. Yet that is what
Craig Barrett of Intel, the world's largest chipmaker, did this week
at an industry conference in Florida. He was only joking, of
course. But his apology for Intel's decision to cancel the next
version of its flagship Pentium 4 chip highlights the latest in a
series of stumbles by the company, which has once again been forced to
follow the lead of its much smaller but increasingly feisty
competitor, Advanced Micro Devices (AMD).
At issue is the best approach to making faster chips. For years, Intel
has steadily increased the clock speed of its processors, the fastest
of which now run at 3.4GHz, or 3.4 billion ticks per second. But it
has now fallen victim to the law of diminishing returns. Although
boosting the clock speed increases performance, it also increases the
power consumption of the chip and the need for cooling. (Some of the
very high-speed PCs used by serious gamers are even water-cooled.)
So, rather than concentrating on clock speed, Intel has decided to
boost the performance of future chips in other ways, such as
increasing the amount of on-board cache memory and, in the coming
years, switching its chips to a multi-core design. This means putting
multiple cores in effect, complete processors into a single
chip. These cores can run more slowly, consume less power and generate
less heat, while collectively providing more processing power than a
single core. Multi-core is a way to achieve additional performance
without turning up clock rates, says Dean McCarron, an industry
analyst at Mercury Research. This idea is not new: IBM, Sun and
Hewlett-Packard already sell high-end computers powered by their own
multi-core chips. But it is only recently that PC software has been
able to exploit multiple processors.
Intel's decision to de-emphasise clock speeds is just the latest
example of how the company has reluctantly ended up following where
AMD has previously= led. (Earlier this year, AMD forced Intel to make
a U-turn in its=20 64-bit-chip strategy.) AMD has long argued that
there is more to=20 performance than clock speed, and gives its chips
model numbers giving some= idea of their power. Its new Athlon 64
4000+ chip, for example, announced=20 this week, runs at 2.4GHz, but
its name implies rough equivalence with a w4GHz Intel chip. Intel is
now adopting similar model numbers.
Having abandoned its obsession with raw speed, Intel is embracing the
multi-core approach with great enthusiasm. Paul Otellini, Intel's
number two, who is expected to take over from Mr Barrett next May,
said last month that he expects 40% of desktop chips sold, and 80% of
server chips, to be multi-core by the end of 2006. The switch to
multi-core is, he says, a sea change in computing and a key inflection
point for the industry .
What does all this mean for Moore's law, the rule of thumb coined by
Gordon Moore, Intel's co-founder, which states that the amount of
computing power available at a given price doubles every 18 months?
For most people, Moore's law manifests itself as a steady increase in
clock speed from one year to the next. The cancellation of the 4GHz
version of the Pentium is Intel's clearest admission yet that clock
speed is no longer the best gauge of processor performance:
henceforth, it will increasingly take a back seat to other
metrics. But the law itself, the death of which has been announced
many times, will live on. Mr Barrett insisted this week that it would
continue to apply for at least another 10-15 years. That is because
multi-core designs mean chips' performance can continue to increase
even if the formerly much-trumpeted clock speed does not.
Copyright 2004 The Economist Newspaper and The Economist
Group. All rights reserved.
*** FAIR USE NOTICE. This message contains copyrighted material the
use of which has not been specifically authorized by the copyright
owner. This Internet discussion group is making it available without
profit to group members who have expressed a prior interest in
receiving the included information in their efforts to advance the
understanding of literary, educational, political, and economic
issues, for non-profit research and educational purposes only. I
believe that this constitutes a 'fair use' of the copyrighted material
as provided for in section 107 of the U.S. Copyright Law. If you wish
to use this copyrighted material for purposes of your own that go
beyond 'fair use,' you must obtain permission from the copyright
owner, in this instance, Economist Newspapers.
For more information go to:
http://www.law.cornell.edu/uscode/17/107.shtml
Marcus Didius Falco (falco_marcus_didius@yahoo.co.uk)