Re: XEROX 1100, PDP-10 performance...

[e@flavors.com (Doug Currie, Flavors Technology, Inc.)]

At 1:15 AM 95.03.16, JasonStowe wrote:
>Hello. During some recent reading of some old comp sci materials, I found
>a
>reference to a "SUMEX, PDP-10, with a KI-10 uniprocessor 256k core memory"
>and a "XEROX 1100 series". I wondered if anyone out there might know how
>many
>million instructions per second (or fractions of MIPS) such machines would
>be
>able to run. Otherwise, if someone knows how fast relative to a
>Pentium90??
>Any comments on these old machines' performance(late 1970's/early 1980's)
>would be of help.

These were microcoded machines. The KI-10 and KL-10 were designed by Bob
Reid, and used in the popular DecSystem 2020. I was lucky to work with Bob
in 1977, right after he did these CPUs, on a new CPU for Gould; he did the
hardware, I did the microcode. Based on what I remember about the
technology of those machines, the microcode cycle was 100 nsec (ECL) to 250
nsec (TTL). A user level instruction might take four to twenty (or more for
floating point or multiply/divide) microcycles. So the top end for these
"fire breathers" was about 1 MIPS even with aggressive pipelining and
caching. The VAX was also being developed at this time. The fastest first
generation Vaxen were also about 1 MIPS.

Gabriel has a book on "Performance and Evaluation of LISP Systems" (MIT
Press?) which will have data on these and many other systems. I don't have
my copy handy, so I can't give you any hard data (and please forgive me if
I've got the title wrong). If you can't find the book, write me directly
for a better reference.

e