Anders Magnusson wrote:
> 
> >
> > 2080 = fastest PDP10 ever made.
> >
> Was it ever made? Wasn't it cancelled before it was finished?
With Ron Bingham as the manager, Pat Sullivan as the memmory guy, Bugs
Mclean and Ted Hess as the micro-coders, Don lewine as (sorry can't
recall what don called himself), Steve Pomfret, Marty Schwartz, and me
on the I/O, bob Elkind and (again a name I can't recall) as the CPU
hardware guys, The Pope as the "TECH", Nate Kerlinovich as Ron's deputy,
and help from all the other 10/20 folks stashed around the company - a
commitment was made to build a 4MIP (2060x4) capable machine.
WIth the cpu guys, the microcoders, and that memory guy, the first set
of simulations showed it to be a little faster (about 8X) when we were
willing to talk about it.
Pat said something along the lines of how fast do you want data from
memory?  And did his magic.
The microcoders/modellers and cpu guys kept finding ways to speed the
machine.  One famous bit was the JRST0 instruction that executed in 0
time.
This was a pipelined implementation.  Lots of micromachines inside.
The public statement, as Allison said, was it cost too much with too
little return.  The truth is a little different.
CPU was tunable via microcode for up to 50 MIPS.
The manufacturing cost, in 1980 or so dollars for the basic config was
47K US. The Venus cost was just over 50K - these are CPU only
comparisons cause we said the periphs kind of balanced out.
We were artificially limited as to "enhancements" to the straight 10/20
stuff.  We were not allowed to extend the addressing on the KC-10 for
the 100K ecl logic we used (original code name Hephastus, changed to
Jupiter when someone found out that Hephastus, as vulcan, ate the
children of venus).  When we were naming the dec name of the cpu, we
asked Ken if we could call it KO-10 but he said no.  He smiled at the
compliment but like the C for hundred.
There were some problems with briefing all of the consulting engineers
all the time on the advantages of pipeline design, and why we could not
turn Venus into a pipeline design.  Bob Elkind was finally detached from
us to move to Venus.  That was when I gave up.  The built machine = the
real one = had to have some microcode changes to balance the I/O, Ted
and Bugs can recall the rest of that story but it basically was one of
time clock running, one set of code had it really running with about
50MIPS of CPU and a Megabyte/second of I/O, we wanted something better
than 25Mips, and 25MB/S I/O over the interfaces - massbus, unibus, and
then the CI bus.  We were precluded from running an IBM Channel emulator
(we had one designed to complement the DX-11, that we called the DY-11 -
Frank Zereski had designed that as a follow on to the DX).

Sorry for forgetting folks names, but that is the truth as I recall it
...
bob 
-- 
real address is shsrms at erols dot com
The opinions expressed here may or may not be mine.