Article 1110 of dec.notes.technology.dechips:
Title: HP: Alpha tops out at 600MHz
Reply Title: Benefits from out-of-order techniques

I have a bit of a unique perspective on this one.  I was at HP
when they shrank their initial multi-chip CMOS implementation
of the PA-RISC architecture to produce the 1-HP-chip/1-TI-chip
pair that was the basis of the 7xx "Snakes" boxes.  (In fact I
can even claim some amount of notoriety for the presence of the
5 operand instructions that crept into the PA-RISC architecture
at that point).  The 700 spawned a number of leveraged designs,
namely the 7100, 7150, 7200 and 7300 chips.

[Interesting historical footnote: the basic integer and address
formation pipeline, including access rights checking, segment
register lookup, etc, was leverage from a TTL MSI breadboard,
through two generations of NMOS, and four generations of CMOS.
In fact while verifying the 7100 chip we discovered a corner
case in the validation of coprocessor instructions with quad-
precision operands that was traced right back to the original
TTL schematics.]

While various teams were grinding out the 7150, 7200 and 7300
chips, the best and brightest of the engineers in Fort Collins
were arguing that the time had truly come to do a clean sheet
new design, leaving all vestiges of the past behind.  They
undertook a number of studies to determine how best to build
a superscalar processor of degree greater than two.  The bottom
line of all their studies was that if all other factors were
held constant they believed they could squeeze 35 to 40% more
performance from an out-of-order implementation.

"What factors?" I hear you ask.  Unfortunately I don't remember
exactly.  Obviously the semiconductor technology.  And also I
believe the die size.  What I do remember is that they remained
steadfastly opposed to implementing an on-chip L1 data cache.

After leaving HP I joined SEG/AD around the time EV45 and EV5
were being designed.  I had hoped to participate in the EV6 design
and tried to generate some interest in out-of-order techniques.
All of the big names told me that Digital would always achieve
performance by eschewing such foolishness.  But ultimately Dirk
Meyer had the courage and intellectual honesty to suggest that
the idea should not be discarded out of hand and that it deserved
at least an investigation.  And guess what... they came up with
about the same number: 35 to 40% more performance over an equi-
valent in-order machine!