From Newsgroup: comp.arch

According to Anton Ertl <anton@mips.complang.tuwien.ac.at>:

EricP <ThatWouldBeTelling@thevillage.com> writes:

I don't know what 780 used for its integer multiplier
(I can't find it in the schematics).

Then I expect that it used microcode to do 1 multiplier bit per cycle.

This paper measured the 780's performance by making microcode histograms
of running programs. In one table it says the integer multiply and divide
were handled by the same subsystem as floating point instructions.

https://people.csail.mit.edu/emer/media/papers/1984.05.isca.vax.pdf

Adding to the confusion, there was an optional floating point accelerator
which presumably added hardware to do what used to be done in microcode,
but they don't mention that at all.
--
Regards,
John Levine, johnl@taugh.com, Primary Perpetrator of "The Internet for Dummies",
Please consider the environment before reading this e-mail. https://jl.ly
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch

On 2026-May-10 15:09, John Levine wrote:

According to Anton Ertl <anton@mips.complang.tuwien.ac.at>:

EricP <ThatWouldBeTelling@thevillage.com> writes:

I don't know what 780 used for its integer multiplier
(I can't find it in the schematics).

Then I expect that it used microcode to do 1 multiplier bit per cycle.

This paper measured the 780's performance by making microcode histograms
of running programs. In one table it says the integer multiply and divide were handled by the same subsystem as floating point instructions.

https://people.csail.mit.edu/emer/media/papers/1984.05.isca.vax.pdf

Adding to the confusion, there was an optional floating point accelerator which presumably added hardware to do what used to be done in microcode,
but they don't mention that at all.

There is a follow-up to that 1984 paper
"A Characterization of Processor Performance in the VAX-11/780"
from 1998 by the same authors called
"RETROSPECTIVE: Characterization of Processor Performance in the VAX-11 /780" which makes that following admission:

"In particular, while the VAX-11/780,
which was introduced in 1978, was probably the
preeminent timesharing machine on university
campuses at that time, very little was known about
how it worked or exactly what its performance
was.

In particular, before 1980, even inside Digital
the fact that some benchmarks ran at less than the
widely-believed 1 MIPS was known to only a very
small number of people. And the fact that on real
multiuser workloads the 11/780 typically executed
instructions at only 0.5 MIPS was apparently
unknown. Furthermore, somewhat embarrassingly,
both facts were unknown to the architects of
some of the successor machines. That meant that
those designs were optimizing to the presumed 5
average CPI of the ll/780, where in fact another 5
cycles per instruction were totally unaccounted for.
It was only following some other measurements by
one of us (Joel) in which a frequency counter was
hooked up to record MIPS and he was shocked to
read 0.5 MHz where he expected 1.0 MHz, that a
more widespread account of the 0.5 MIPS rating
was propagated. Still, so widely believed was the
1.0 MIPS number, in fact, that one of our ISCA referees
didnrCOt believe the data, making the rCLmandatoryrCY
recommendation to rCLexplain why Table 8
and 1st bullet, pg. 23, seem to imply average VAX
780 instruction takes > 2 us; should be ~1 us.rCY
"

And thus was born the marketing term VUP: VAX Unit of Processing
where 1 VUP was defined as the speed of a 780.
And all was right with the world again,
or at least the DEC marketing dept.

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch

According to EricP <ThatWouldBeTelling@thevillage.com>:

There is a follow-up to that 1984 paper
"A Characterization of Processor Performance in the VAX-11/780"
from 1998 by the same authors called
"RETROSPECTIVE: Characterization of Processor Performance in the VAX-11 /780" >which makes that following admission:

"In particular, while the VAX-11/780,
which was introduced in 1978, was probably the
preeminent timesharing machine on university
campuses at that time, very little was known about
how it worked or exactly what its performance
was.

In particular, before 1980, even inside Digital
the fact that some benchmarks ran at less than the
widely-believed 1 MIPS was known to only a very
small number of people. And the fact that on real
multiuser workloads the 11/780 typically executed
instructions at only 0.5 MIPS was apparently
unknown.

I saw that and was kind of surprised. I don't know when I first heard this, but it was widely believed that the 780 was about the same speed at a 370/158 which IBM said was 1 MIPS. The obvious implication is that VAX did twice as much work per instruction as a 370 which is not implausible. So depending on how you define MIPS, it's not strictly false.

How did I know that and people at DEC not? Maybe they needed to get out more. --
Regards,
John Levine, johnl@taugh.com, Primary Perpetrator of "The Internet for Dummies",
Please consider the environment before reading this e-mail. https://jl.ly
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch


John Levine <johnl@taugh.com> posted:

According to EricP <ThatWouldBeTelling@thevillage.com>:

There is a follow-up to that 1984 paper
"A Characterization of Processor Performance in the VAX-11/780"
from 1998 by the same authors called
"RETROSPECTIVE: Characterization of Processor Performance in the VAX-11 /780"
which makes that following admission:

"In particular, while the VAX-11/780,
which was introduced in 1978, was probably the
preeminent timesharing machine on university
campuses at that time, very little was known about
how it worked or exactly what its performance
was.

In particular, before 1980, even inside Digital
the fact that some benchmarks ran at less than the
widely-believed 1 MIPS was known to only a very
small number of people. And the fact that on real
multiuser workloads the 11/780 typically executed
instructions at only 0.5 MIPS was apparently
unknown.

I saw that and was kind of surprised. I don't know when I first heard this,

I heard this in 1983--as a reference point.

but
it was widely believed that the 780 was about the same speed at a 370/158 which
IBM said was 1 MIPS. The obvious implication is that VAX did twice as much work
per instruction as a 370 which is not implausible. So depending on how you define MIPS, it's not strictly false.

How did I know that and people at DEC not? Maybe they needed to get out more.

Riding high on their laurels.

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.arch

John Levine <johnl@taugh.com> schrieb:

According to EricP <ThatWouldBeTelling@thevillage.com>:

There is a follow-up to that 1984 paper
"A Characterization of Processor Performance in the VAX-11/780"
from 1998 by the same authors called
"RETROSPECTIVE: Characterization of Processor Performance in the VAX-11 /780" >>which makes that following admission:

"In particular, while the VAX-11/780,
which was introduced in 1978, was probably the
preeminent timesharing machine on university
campuses at that time, very little was known about
how it worked or exactly what its performance
was.

In particular, before 1980, even inside Digital
the fact that some benchmarks ran at less than the
widely-believed 1 MIPS was known to only a very
small number of people. And the fact that on real
multiuser workloads the 11/780 typically executed
instructions at only 0.5 MIPS was apparently
unknown.

I saw that and was kind of surprised. I don't know when I first heard this, but
it was widely believed that the 780 was about the same speed at a 370/158 which
IBM said was 1 MIPS. The obvious implication is that VAX did twice as much work
per instruction as a 370 which is not implausible. So depending on how you define MIPS, it's not strictly false.

There's a reason why MIPS has the backronym "Meaningless Indicator
of Processor Speed".

How did I know that and people at DEC not? Maybe they needed to get out more.

Maybe they were ignorant about the typical workload of their
machines?
--
This USENET posting was made without artificial intelligence,
artificial impertinence, artificial arrogance, artificial stupidity,
artificial flavorings or artificial colorants.
--- Synchronet 3.22a-Linux NewsLink 1.2