Analysis: Speaking of Silicon
What Intel really builds
By Ron Wilson, EETimes
Every time Intel announces a new Pentium CPU there is a lengthy debate about
its relative performance. This has become a sort of ritual flogging, in which
each advocate of each RISC architecture in turn stands up to demonstrate
how his current favorite is 30 percent faster than Intel.
I'm sure this provides some solace for the supporters of architectures that
have been rendered irrelevant on the desktop. But it misses an important
point that might be worth restating.
That point is that Intel is no longer really designing microprocessors. These
days, what Intel designs is a complex system including at least one CMOS
process, a generation of fab equipment, several near-billion-dollar fabs,
a microprocessor, a set of supporting chips and a motherboard. If all the
pieces aren't ready on time, the microprocessor by itself will have little
value.
Nor is this a quibble over terms. No other microprocessor vendor is in this
position. Given the comparatively minuscule volumes and enormous system-cost
budgets other desktop engines enjoy, none of them needs to worry over-much
about these issues.
But Intel must trade back and forth between a process that has to go almost
at once into high-volume production, a die size that must meet
corporate-contribution goals, yield levels that cannot be allowed to wander,
an architecture that must represent a significant improvement over the best
clock rates of the previous generation, core logic that must be understandable
yet must exploit the CPU, and a motherboard design accessible to a large
number of board suppliers with a wide range of skills.
Compare that situation with the easy life of the other desktops. Alpha, to
pick an example, is generally the performance leader of the month by
Microprocessor Forum. Digital Equipment Corp. accomplishes this with increasingly
sophisticated architecture, good CMOS processes and, more than anything else,
enormous clock rates.
But for those hundreds of megahertz, Digital trades off yield and usability.
The company has been known to announce a CPU when there was only one die
working at the claimed speed. And, with no external customer using the fast
CPUs in significant volumes, only one design team need endure the challenges
of realizing that speed on a real-life motherboard.
This is not a criticism of Digital. Its strategy is well-suited to its market
needs. But it makes clear the folly of any direct speed comparison between
the latest Alpha CPU--or even the latest x86 look-alike--and the latest Pentium.
The wonder accomplished repeatedly by Intel's design teams is not that they
produce the fastest desktop CPUs. It is that, given their constraints, they
actually come close.
(c) 1997 CMP Media, Inc
[This article comes from EE Times in a joint cooperative effort
with the Motley Fool. For more articles like it, please look at Fool's Gold
every weekend or simply go to the Fool's Gold Mine and page through our back
issues, which all have clever and cool EE Times articles in
them.]
|