Glen has asked me to help him accept this wonderful award, since the
stroke he suffered seven years ago makes speech rather difficult for
him.  You can tell from the sparkle in his eye that he is quite
pleased by the whole thing.  If he were able to give this speech, I'm
sure that he would have offered gratious thanks to those that made it
possible and share with us many drops of wisdom stemming from his
preceptive eye and his unique outlook.  So I'll do the best I can.

First, let me thank the IEEE Computer Society, the SC2000 organizers,
and the awards committee, on his behalf, for bringing him this
opportunity.  A special thanks to the people who added their support
for his nomination, including George Michaels - the man who got this
whole conference going, Doug Englebart - the inventor of the mouse and
windows and an old friend, Yale Patt - who's undergraduate time with
Glen shaped his career, Chuck House, and Burton Smith, David Wood, and
Greg Papadopoulos - a cadre of fine architects.

Before finishing our thanks, I would like to share with you a
"next-best" to the thoughts that Glen would impart if he could - some
of the thoughts and outlooks that fell out as part of growing up with
this man.

I know that this award has made Glen very happy.  It shows in his walk
and in his talk - maybe not speedy, but faster than a few weeks ago.
It happened, as well, when he went to Washington to receive the
National Medal of Technology last winter.  But in some ways, I think
this award has a deeper connection for Glen.  There are many parallels
between Glen and Seymour Cray: contemporaries, competitors perhaps, from
small towns in the midwest, coming to professional life upon the first
wave of computing.  Two keen architectural visions.

In 1976 two breakthrough machines were introduced.  The Cray-1,
delivering 20 MFlops for about 10M$.  The other was the Floating-Point
Systems AP-120B, delivering about 3 MFLOPS for 50K$.  It was often
called, the "Poor man's Cray".  For many of you, the Cray was what you
shared in the national computer centers, the FPS box was what you
connected to the minicomputer in your lab.  It would be fifteen years
before that price performance point was matched by conventional
designs.  Where the Cray introduced the lean pipeline and instruction
sets that would eventually be coined RISC, while pushing circuit
technology, the FPS design employed vanilla technology but exposed
parallelism at the instruction set - in what would eventually be
called VLIW and most recently EPIC in the IA64. 

So let me take you backwards a fowards a bit from that point.  I
belive Glen's outlook on computing was rooted in experience as a young
mathematician running, rather programming, rooms full of people with
hand computers at what is now Lawrence Livermore National Lab to solve
various differential equations and integral equations presented by the
physicists.  This was with Sid Fernbach and some of the founding folks
here.  Is Hans Bruijnes here?  This was explicit parallel execution,
if there ever was.  He would explain to me that the sequence of
operations would go down the page and data points across.  Some people
like to pick one data point and take it through the steps.  Others
would apply one step to each data point.  Of course, when you automate
this as a wide instruction loop - as was rediscovered 30 years later
as "software pipelining" you work on the diagonal - so you do a
different step of each of a set of data points simultaneously.

The deeper message was that computers needed to be involved in the
process of figuring out how to solve scientific problems, not just in
grinding out the numbers once you have figured out exactly how to do
it.  When he returned to graduate school at UCLA, he also took a
job as director of computing at Ramo-Worldridge and continued with
them after taking a position at UCSB.  They had a big expensive
computer - the RW-400 with all of about a kiloword of memory - but
more importantly an large display tube.  He built on that an
interactive mathematics system - very much like matlab today - where
you could associate data arrays with variables.  Manipulate them with
individual keystrokes, including graphing them on the display.  One of
the patents on that was the association of mathematical functions to
keys - which years later was the core of TRW's lawsuit with HP over
the programmable calculator.  This "on-line system" was very exciting
to young physicists, like Screifert and Feynman, before they became
nobel laureates.  This clearly a theme in his work - computers should
provide a means to help you develop intuition, help you think about
solving new problems, not just solve old ones faster!

He returned to UCSB in 61 and took the on-line system forward to
create an interactive computer classroom for teaching calculus.  Of
course, no displays existed so they had to build them out of
oscilloscope tubes.  And to get those function keys they needed to
build keyboards.  And of course, timesharing hadn't really been
invented yet, so they developed a timesharing system that would give
16 users 100ms response time - all with a few kilobytes of memory.
Everytime the drum went around it would bring in an entire user
machine state.  Burton would call that multithreading, I guess. They
got funding to rebuild the whole system on an IBM mainframe, which he
managed to do while running the project from a hospital bedroom for a
year.  They networked the system to classrooms in several other
departments, for chemistry and physics research, and soon to other
campuses, including Harvard, making it one of the first networked
computers.  It was the reason that UCSB was one of the original four
ARPAnet sites.  I think the lesson he would draw from all this is that
if you have something really important to accomplish, don't let a few
technological hurdles stand in your way, much less a few people who
say it cannot be done.  The passion he felt for his work - getting off
a two or three in the morning seven days a week - was inspiring to
many.  I think also bringing inquisitive young minds into his world
was so crucial to him - teaching and research could not be divorced.

He used to tell me about a consulting gig at IBM he had somewhere in
this timeframe.  It turned out to be a discussion with Watson on the
future of computing - the other consultants were Courant, Alverez, and
Norbert Weiner.

Although graphical visualization was nice, Glen was evidently still
frustrated by the problem solving bandwidth between himself and the
computer.  He never was much of a typist - that probably why he put
all those mathematical operators on individual keys - and he would
talk out loud with every key stroke.  Clearly, these machines out to
understand speech.  He had lots of ideas about how to analyze and
synthesize speech, but in the 60s the machines were nowhere near
powerful enough to pull it off.  That's how he got into computer
architecture. Left the university to do a startup (it wasn't so common
in 69) and build high performance computers capable of speech
processing.  Where did he get the performance?  Those same techniques
that he used with all those hand computer operators - harness the
parallelism.  In 1973 the AP90-B was operational.  It had multiple
operations per instruction - addition and multiplication along with
address calculation and memory access.  It used block floating point,
because the engineering associated with full floating point was
prohibitive.  The machine was closely couple to a "Macro Processor",
which ran the interactive timesharing system, through a message
passing interface.  This design was license to a company that made
floating point units, (remember when you could plug them in) and
became the FPS AP120B. 

Another parallel with Seymore Cray - grossly underestimate the
significance of your work.  The original licensing of the
AP design to FPS assumed that a maximum of 50 systems would ever be
built. 

Although the AP took off in many traditional areas of scientific
computing, he was very much occupied with digital speech - analysis,
synthesis, storage, transmission.  I can remember being dragged out of
bed way too early on weekend mornings because he wanted to test out his
newest algorithms on a vocal tract that hadn't changed yet.  One time
a friend happened to be spending the night so he got dragged along
too.  He tried so hard afterward to explain to people that he had met
a computer that talked to him.

Glen built a number of systems around the FPS box, including large
scale particle simulation systems, and somehow managed to keep his
company alive without going explonential in either direction. I cut my
programming teeth on the AP when I went to work for him in high
school.  Glen is certainly a man who believes in teaching the
fundamentals.  On day one he handed me a TTL cookbook and said, study
these chips, then will talk about how to make a computer out of them.
My older brother Randy worked for him as an operator running the
on-line system on campus.  He says, "You got off easy!"  Glen started
him out by handing him a listing of the operating system and pointing
to a line in the code - start here!  Glen clearly believed in starting
from foundations.

Glen developed a number of increasingly compact systems for doing
real-time speech processing, leading to the first full scale
single-chip signal processing microprocessor in 1982.  This was a
joint project with Motorola.  Motorola retained military rights,
Culler had commercial rights.  It was done in an experiment platinum
silicide CMOS technology - to support the extensive busses and wide
datapaths - which never became cost effective for commercial use.  It
was several years before the signal processing chips from TI and
Motorola move up to this level.  Although it wasn't a commercial
success, some of the first digital music recordings and digital
transmission of speech - which dominates cellular telephony took place
in his lab.  Many of the techniques he was using were rediscovered in
the recent wavelets work.  I had the chance to work with him a bit
during this time as a logic design.  The quote I remember most was,
"The first thing you learn in this business is how to take everything
you've done and start over with a better idea."

One of the really unusual things he did with this technology was an
invention called the chromophone.  The idea here was to map the
fundamental characteristics of speech into the visible spectrum on a
television set so that a deaf person might be able to see the patterns
of sound.  This wasn't the narrow validation that an individual
phoneme was correctly pronounced - it was color and intensity and
change.  Perhaps a person would come to recognize the pattern of the
sound of a door closing.  The message there was tremendous faith in
the human ability to extract meaning and understand from what they
perceive. 

In 1980, I went to work at Livermore developing the Cray TimeSharing
System (CTSS).  We would have long arguments.  Glen thought it was
just terrible that Cray had put all that wonderful parallelism in his
machines, but then put instruction sets in place that tried to create
the illusion that everything was sequential.  I don't believe the two
men ever met, but I would have loved to hear the discussion at that
backyard picnic.

Through the 80s, his company rose with the tide of the minisuper era
and the venture capital that came with it.  He developed the Personal
Supercomputer, which sat side-by-side with the Sun3 workstation, and
the Unix Multiprocessor computer server.  One of the really exciting
aspects of this machine was the compiler technology, which you take
the critical loops and automatically generate VLIW microcode - like we
used to do by hand in the 70s.  They were selling this as a product
before software pipelining became a dissertation topic.  Of course,
that era was brought down by the sheer technological horsepower in the
32-bit CMOS RISC microprocessors - and his company went the way of
multiflow, cydrome, convex, SCS, and others.  He continued as Chief
Architect for Star Technologies - producing the first sparc-based
vector processor.  He was well along in the development of general
design with general, decoupled address generators feeding a dataflow
pipeline, when his stoke cut short his work in the early 90s.

So, now that Glen has time to look back and forward on the role of
computing, I'm sure that he would have many inciteful thoughts,
besides these remembrances of mine - but I'm sure he would want to
close with very deep thanks to the many people that worked with him to
bring his vision to reality, to his family that fully embraced his
creative spirit, and above all dedicate this award to Susi, his wife
of 47 years who passed away in November, three years ago.