[PLUG] Re: IEEE Speaker: The Pentium Chronicles

Keith Lofstrom keithl at kl-ic.com
Sat Apr 23 02:35:58 UTC 2005


Last night was an amazing talk by Bob Colwell, chief architect for
the Intel P6, which was the grandfather of most of the current 
family of Intel processors.  So arguably this fellow designed most
of the world's computing iron - along with a few thousand friends.

There were more people there than I have ever seen at a local IEEE
presentation - 360 registered, more snuck in, and many turned away
at the door.  The talk was long and a bit disorganized, but there
was a great deal of information about the process that decides what
goes into the processor and what does not.  The politics involved 
are enormous - those architectural decisions decided how a hundred
billion dollars worth of computer systems work.  Telling Compaq, 
for example, that the level 2 cache goes in the package, eliminating
one of Compaq's chief differentiators and possibly leading to it's
weakening and absorption by HP.  Telling a Cray chief architect
"sorry, no hidden instructions for you, unless you are willing to
buy 200 million processors and your own design".  Deciding to bet
the company on out-of-order instruction sequencing, and remaining
loyal to older users by keeping legacy instructions (if they had
remembered to do that with Itanium, then Opteron would not have 
cleaned their clock).

The audience was about half Intel; there was a lot of Intel politics
discussed.  One interesting factoid was about the FDIV floating point
divide bug.  There were about a dozen engineers involved in the
various stages of designing and testing the hardware with that bug;
while it cost Intel half a billion dollars to fix, the engineers were
not publically censured.  Colwell does not know their names, but knows
they still have viable careers at Intel.  The idea is that when you 
take risks, you risk mistakes;  if you punish mistakes, you punish
risk taking, which is lethal for a high-tech company.  The approach
they take instead is to plan to manage risk and the inevitable mistakes
that occur.  Unlike the Beast of Redmond, Intel ultimately owned up to
the mistake and fixed their customers machines.

Intel has a reputation as a hard-ass company, but I was impressed by
their having a ghost of a clue, rare for a company that size and that
strongly driven.  Colwell (now a consultant) is a hero to most of the
Intel engineers in that room.

Keith

-- 
Keith Lofstrom          keithl at keithl.com         Voice (503)-520-1993
KLIC --- Keith Lofstrom Integrated Circuits --- "Your Ideas in Silicon"
Design Contracting in Bipolar and CMOS - Analog, Digital, and Scan ICs



More information about the PLUG mailing list