CMM in outside software (was Re: Ada and QNX)

["Robert I. Eachus" <rieachus@earthlink.net>]

(This grew from an endorsement of Robert Dewar's idea into a rant on the
shortsightedness of American corporate management, in the guise of a
discussion of designing CPU chips.)

Robert Dewar wrote:
> 
> What arguments would impress management. Well try this on
> for size "Ada will help you achieve level 3 CMM more rapidly
> and realiably". I suspect this is definitely true, and I could
> (and have) written papers to support this, but I would be
> interested in other people's viewpoints. There's a tecnical
> advantage that definitely WOULD play in some circles.

I'm not sure, and I am not sure that there are many managements that
understand what a level 3 CMM will do for them other than check off some
government contracting box.

But take a look at Intel and AMD right now.  Intel has put a lot of
effort into getting a chip design and manufacturing capability that is
repeatable.  By doing so they have become one of the largest companies
in the world.  And take my word for it, Intel is better at designing
chips--not architectures--than almost anyone else in the world.

Up until a year ago, Intel used AMD as a footstool, and kicked them from
time to time when Intel thought AMD was getting uppity.  Then came the
Athlon.  The Athlon didn't come out of nowhere.  The design team that
developed it previously developed the K6 and K6/2 processors.  A few
months before the Athlon was announced, an Intel executive talked about
how in 4Q2000, Intel would be introducing the Willamette at 1.1 GHz, and
the fastest Athlon would be running at 666 MHz.  AMD introduced the
Athlon at four speeds, 500, 550, 600, and 650 MHz.  Ooops!  Intel's
fastest processor at the time was running at 600 MHz.  Wouldn't have
mattered, except that the Athlon was faster per clock at floating-point,
and had a much faster memory bus, had much bigger L1 caches, etc.  Even
at the same clock speed it was a much better chip, all around.  AMD then
took this new architecture, and introduced the K75 which was the first 1
GHz chip, the Thunderbird, with on-chip L2 cache, and a little brother
the Spitfire (Duron), that walked all over the Celeron.  (Right now, AMD
is not competing with the Celerons--they have repositioned the Duron
against the Pentium III.)

AMD this week announced a second 1.2 Ghz chip with a 266 MHz Front side
bus, a chipset to allow DDR (double data rate) SDRAM to be used with the
Athlon and Duron, and missed their first scheduled product introduction
this year.  They decided to wait a week or so and announce the newest
Athlon core, the Mustang, at Comdex.  (The chip was initially scheduled
for an October release.)

What has all this got to do with the SEI CMM?  Well, after twenty years
in the wilderness, AMD finally realized that what they needed was not a
better product than Intel, they needed a better process for developing
new products.  The "Classic" Athlon (at 0.25 micron) was released about
the same time as Intel's new Coppermine (0.18 core) for the Pentium
III.  Then came the K75 (Pluto) core (at 0.18 micron), the Thunderbird
(again 0.18 but some with copper interconnects), and Mustang (again 0.18
copper) is the first of the next generation of AMD chips to be
announced, and will be announced next week.  A week or to later Intel
will announce their new Pentium chip, the Willamette.

So AMD will have gone through three new architecture releases in the
time it takes Intel to do one.  But that is not exactly true.  Intel has
more than one product design team, and expects a major new design like
the Willamette to take up to four years.  Unfortunately, Willamette took
longer than that.  (And Merced/Itanium has taken so long that it is
obsolete before it is released. Intel is hoping that McKinley will
change that at some point next year.)

Ah, you say, the K75 and Thunderbird were just minor tweaks.  In one
sense true, in another irrelevant. The difference between Katmai and
Coppermine is about the same as the difference between the K75 and the
T-bird.  The Mustang is probably a bit more of a change from the T-bird,
but the next core after the Mustang is the Sledgehammer, and it is at
least as innovative as the original Pentium Pro.  Of course, the
Sledgehammer is not on schedule for a late 2001 release--it is a little
early since it taped out this week.  Between Mustang and Sledgehammer,
there will be the Palomino desktop chip, and a mobile variant, and the
Morgan (new Duron) and its mobile version. That's right, AMD is in the
midst of announcing more major new products in the October to February
time period than ever before in their history, so they cleaned up the
next major project early and got it out of the way.  (AMD may or may not
also release the Clawhammer somewhere in there. Clawhammer is a
prototype x86-64 chip that may be in house only. But I heard rumors they
might merge the ISA enhansements from the Clawhammer into the Palomino. 
How's that for reuse.)
 
Intel is hoping to catch up to AMD next year.  But they haven't gotten
the message yet.  Right now there is a lot of talk about whether the
Willamette will be able to match the Athlon when it is released, or if
they will have to wait for the 1.7 GHz or 2.0 GHz versions.  But AMD
will have a new (and from what I have heard) much faster chip next
week.  The Willamette may not even hold the clock speed record when
shipped.  Jerry Saunders' (AMD's CEO) last word on the subject was new
higher speeds every few weeks, 1.5 GHz by January.  Right now Willamette
is expected to be announced at 1.4 and 1.5 GHz on Nov. 20th.  The
Mustang is expected to be announced at 1.4 GHz and several other speeds
next week.  The speculation is on the "several other speeds. ;-)

How many billions of dollars has working on their development process
been worth to AMD?  (My guess is about 1 billion cash, and 10 billion in
market cap so far.)  And how much has lack of a better process cost
Intel.  (Not that much so far, maybe 100 billion of market cap or so,
but wait 'til next year.)  Who else has been focusing on process
recently?  Well, Jack Welsh has been doing it for years at GE.  Six
Sigma is almost a corporate religion now.  And it really is the case,
when some process doesn't work right, they don't brush it off, they look
at what can be done to fix the process--if it really is broken.

Given all that, wouldn't you think that more company executives would
focus their energies on improving internal processes?