Performance of PDP-10 vs -11 from the factory

[D R]

I know Oscar's -10 and -11 kits are using a Rasp Pi driven simulator to make like the old machines.

I never had access to a -10 or until later, an -11.  The -8 was our toy at the University Computing Center.

If  you had a real DEC  until, with a -10 and and -11 sitting side by side, which would run faster?  Were there add on modules that would ramp up the performance?  I think  8088 had a module later which if the motherboard was right,  you could drop in the math chip to make things doing floating point go much faster.  Was there such for a DEC unit?

[Steven A. Falco]

I don't think there is any way to answer that. Firstly, there are many different I/O devices. So which disk or tape unit(s) have you bought? Similarly, how much memory did you buy and is it core or semiconductor?

And then there is the question of software. Are you running Fortran code? Assembler code?

And of course which hardware CPU? There were many different PDP-11 CPUs, and also a number of PDP-10 CPUs.

So basically I think you could make the numbers come out however you'd like. :-)

As to accelerators, there were several. Floating point could be optional, for example.

Steve

> --
> You received this message because you are subscribed to the Google Groups "[PiDP-11]" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to pidp-11+u...@googlegroups.com <mailto:pidp-11+u...@googlegroups.com>.
> To view this discussion visit https://groups.google.com/d/msgid/pidp-11/2c019761-d985-402e-b8d1-6ad14c45f05dn%40googlegroups.com <https://groups.google.com/d/msgid/pidp-11/2c019761-d985-402e-b8d1-6ad14c45f05dn%40googlegroups.com?utm_medium=email&utm_source=footer>.

[Bradford Miller]

You might find the following table helpful

On 3/20/25 04:21 PM, D R wrote:

I know Oscar's -10 and -11 kits are using a Rasp Pi driven simulator to make like the old machines.
I never had access to a -10 or until later, an -11.  The -8 was our toy at the University Computing Center.
If  you had a real DEC  until, with a -10 and and -11 sitting side by side, which would run faster?  Were there add on modules that would ramp up the performance?  I think  8088 had a module later which if the motherboard was right,  you could drop in the math chip to make things doing floating point go much faster.  Was there such for a DEC unit?
--
You received this message because you are subscribed to the Google Groups "[PiDP-11]" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pidp-11+u...@googlegroups.com <mailto:pidp-11+u...@googlegroups.com>.
To view this discussion visit https://groups.google.com/d/msgid/pidp-11/2c019761-d985-402e-b8d1-6ad14c45f05dn%40googlegroups.com <https://groups.google.com/d/msgid/pidp-11/2c019761-d985-402e-b8d1-6ad14c45f05dn%40googlegroups.com?utm_medium=email&utm_source=footer>.

--
You received this message because you are subscribed to the Google Groups "[PiDP-11]" group.

To unsubscribe from this group and stop receiving emails from it, send an email to pidp-11+u...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/pidp-11/94aed9ba-b119-401d-ba4a-0641527f8771%40gmail.com.

[Charles Ess]

Specifically, there was an "FP-11-F Floating-Point Processor" that would be roughly equivalent to the math chips for the 8088 and 80286:

<https://gunkies.org/wiki/FP11-F_Floating-Point_Processor>
I saw an ad for it that I can no longer find detailing how much it improved calculations: if it crops up, I'll pass it along.

I'm guessing this is a fairly complete list (from 1994) of add- ons:

<https://groups.google.com/g/vmsnet.pdp-11/c/1BLniOPv6D4>

I'm also guessing that no one has decided they need to code the emulator of the FP11-F that could plug into the PDP11 as simulated in simh and run on the Pi ...

Have fun in all events,

- c.

[Three Jeeps]

From the DEC PDP8 and PDP11 architectures, there were various add-ons to improve performance.  In no particular order:

EIS - Extended instruction set, a chip that the execution of additional instructions

Floating Point - an option that was a board consisting of hardware that did floating point computations.

Cache - a add in cache board

Boards that supported DMA. While not specifically increasing processor performance per se, overall throughput was increased.

In the architectures of the 90's 2000s etc these board architectures were added to the actual CPU chips.

The way to evaluate performance of the DEC pdp8 and pdp11 would be through benchmarks....a whole nother discussion of the types and how to evaluate.

The performance evaluation is couched in the bread of the types of benchmarks. DEC had their own variation of performance benchmarks (VPU, Vax Performance Units)

The architectures of the pdp 8 and pdp11s are typical CISC and modern architectures are based on the RISC model...again, a whole nother discussion.

There also is a difference between CPU performance (instructions per clock) (the whole RISC vs CISC comparisons, and various other CPU architectures)  versus system performance/throughput.

[Johnny Billquist]

You're trying to compare apples and oranges...

Johnny

On 2025-03-20 21:21, D R wrote:
> I know Oscar's -10 and -11 kits are using a Rasp Pi driven simulator to
> make like the old machines.
>
> I never had access to a -10 or until later, an -11.  The -8 was our toy
> at the University Computing Center.
>

> If  you had a real DEC  until, with a -10 and and -11 sittingside by

> side, which would run faster?  Were there add on modules that would ramp
> up the performance?  I think  8088 had a module later which if the
> motherboard was right,  you could drop in the math chip to make things
> doing floating point go much faster.  Was there such for a DEC unit?
>
>

> --
> You received this message because you are subscribed to the Google
> Groups "[PiDP-11]" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to pidp-11+u...@googlegroups.com
> <mailto:pidp-11+u...@googlegroups.com>.
> To view this discussion visit https://groups.google.com/d/msgid/
> pidp-11/2c019761-d985-402e-b8d1-6ad14c45f05dn%40googlegroups.com
> <https://groups.google.com/d/msgid/pidp-11/2c019761-d985-402e-
> b8d1-6ad14c45f05dn%40googlegroups.com?utm_medium=email&utm_source=footer>.

--

Johnny Billquist || "I'm on a bus
|| on a psychedelic trip
email: b...@softjar.se || Reading murder books
pdp is alive! || tryin' to stay hip" - B. Idol

[Three Jeeps]

forgot to add the hardware memory management board that were also an option.

the remapping of memory space (virtual) can have performance impacts.

On Thursday, March 20, 2025 at 4:21:46 PM UTC-4 daleea...@gmail.com wrote:

[Johnny Billquist]

Uh... Depending on which model you are talking about, MMU, as well as FP
and EIS was standard. And depending on model, EIS, as well as MMU and
FPP might be a chip, or a whole board. Same story with cache...
It very much depends on which CPU we're talking about.

And no, the MMU does not affect performance. And pretty much any disk
controller, as well as a bunch of other controllers use DMA, and while
some controllers might have had a processor on board, others did not,
but could still do DMA. And you'll still have to throw a bunch of
external chips and stuff onto a modern CPU if you want to have 64
terminal lines (if that is even possible on a modern machine).

Not sure exactly where you have been looking information up, but you
should read through some processor handbooks, I think.

Anyway, the PDP-8 is very much RISC-like, while the PDP-11 is very CISC,
and so is the PDP-10. However, while the PDP-8 is a 12-bit processor,
the PDP-11 is a 16-bit processor, and the PDP-10 is a 36 bit processor.
And it's really hard to make speed comparisons between a PDP-11 and PDP-10.

And no, performance on DEC machines in general was not measured in VUPs.
That was something DEC only ever did for VAX machines.

Johnny

> for a DEC unit?
>
>
> --
> You received this message because you are subscribed to the Google
> Groups "[PiDP-11]" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to pidp-11+u...@googlegroups.com
> <mailto:pidp-11+u...@googlegroups.com>.
> To view this discussion visit https://groups.google.com/d/msgid/pidp-11/

> f418f483-a9ba-4e07-bb7c-ee97d16f3094n%40googlegroups.com <https://
> groups.google.com/d/msgid/pidp-11/f418f483-a9ba-4e07-bb7c-
> ee97d16f3094n%40googlegroups.com?utm_medium=email&utm_source=footer>.

[timr...@gmail.com]

Very hard to compare architectures that are vastly different.  Even if you get some "so many instructions per second" or MIPS, etc.  The 10 can work on 36 bit words where the 11 is 16.  All depends on what you are doing.  One may be better at one task and the other at another.

[DR]

You are getting to the heart of my question, then.


My experience with computers of that era were the CDC 3600 at the U
which rain mainly physics jobs.  Word length and architecture seemed to
make that a choice for those needing computer resources.


The other machine was a Univac 1108, which was general purpose to some
degree as I  understood it.  The academic branch seemed to use it to
teach languages, programming techniques, etc.  I guess that scientific
purposes would be served, but the majority of output of that type were
tapes created to drive a CalComp plotter and do print jobs.   I don't
recall starting jobs from the physicists, at least until the CDC3600 got
blown  up by the four anti-war guys.


The B5500 was in the corner, almost never went down, and the operator on
that console was sure to bring lots of reading materials or classwork to
occupy themselves.  I think it was oriented towards business needs, but
it was hard to know.  I'm not even sure what types of jobs or languages
it supported.  All I recall is hearing that it was very good at what it
did.  We did not load academic (language training, etc.) onto that machine.


There were a few IBM 360s scattered around, but not sure if they did
anything other that statistics in some  of the social sciences areas in
their buildings.


The PDP-8 was brought in to have a more hands-on machine.  Most of the
time we got  to play with FOCAL and other things  on it, more in support
of language learning.


So the question then was a general orientation that DEC designed each
machine for?  General computing? More scientific analysis? Long word
length to get precision out of each instruction?


Since then there seems to be the concept that a computer is more general
purpose, or at least what I perceive.


And then back to the original question, if I had a 'stock' -10 and a
stock -11 sitting on the floor, and had a FORTRAN program, could both
run it, and  if so would they run with equal speed or alacrity for the
mission?  Would one be optimized to run LISP or SNOBOL or PASCAL or
FORTRAN, etc?  COBOL (that's one that the B5 seemed to run, along with a
few of the 360s in other parts of the campus.)

[DR]

Thank you.  there is a lot to learn and I appreciate the comments from
all who were elbow deep and lived and breathed these machines!

[DR]

Ah, now it is beginning to make more sense to  me.

There are a lot of  years gone by and my question was to help me
understand the whole process.

I did not appreciate that the 10 or 11 wasn't just a souped up version
of the other one.


This discussion helps.  Thanks.  And yes, depending on the peripherals
hung on the thing, it might do a lot of things faster.  Our 1108 had
mainly tape drives (I wish I had one to watch at it mounts and runs
tapes, it was mesmerizing to watch the vacuum columns do their thing),
and a FASTRAN which if I recall one of the engineers saying it was sort 
of like a big garbage can on its side, spinning really fast with a lot
of heads flying over a very smooth surface) and a few dial in phone
lines to do some interactive things.


The idea that some of DEC machines did the same thing would indicate
that the interactive (CompuServ) applications would need to  be
supported by certain hardware, in addition to the basic things that
computers do, CPU functions and I/O functions, vs. pure number crunching
like computers on site at the physics experiments gathering data and
later reducing those data.


I had the pleasure to spend some time with a guy who worked with Seymour
Cray on the original, the XMP and another version.  He talked about the
fact that the Crays were more oriented towards vector calculations which
would be  of little use for a teletype connection, but helpful for
weather and aerodynamic simulation. He had a list of 'tricks' which were
advancements that made sense once he mentioned them, but someone ha to
think of it.  Like having four sets of read/write heads located around a
disk so that you could grab the next part of the record without having
to wait for it to rotate all the way around again.  Like having
instruction pre-fetch which didn't always pan out but if the info you
needed was the next thing in line,  you had it; if not  you had to
figure out where the jump went to.

All this history is interesting to me, and it sounds as if many of the
forum participants were indeed on the leading edge of discovery and
innovation, too. Not only hardware, but the software aspects of it.


Fun overall.  Thanks  Dale

[Johnny Billquist]

Most computers would be classified as "general purpose". In theory,
anything that can be done on any computer can be done on any other
computer. I don't think there anything that can be called a computer
which isn't turing complete.

With that said, if we talk specifics on a few DEC computers then.

The PDP-8 was designed as a cheap, small computer for all kind of use
back in the 60s. It's easy to program, and there are compilers and
interpreters for a fair number of languages. But it certainly have some
limitations. First of all, with a word length of 12 bits, the range for
arithmetic isn't that great unless you start doing multiple words, but
that requires more memory for the code, and memory space is also limited
by the 12-bit word size. Basically, you can address 4K word (or 12 bits)
somewhat reasonably, although, being a very RISC-like machine, even
addressing 4K is somewhat more work. The architecture have a very
limited instruction set, and very few registers, meaning it can easily
become quite a lot of code to do some work, which again works against it
with the limited amount of memory. There exists hardware extensions to
go to 32K word address space, but that is somewhat more code to make use
of (there was even a 128K option). There also were hardware extensions
to do a bit more arithmetic by hardware to accelerate things, and there
exists an external floating point processor if you are interested in
floating point performance. The primary programming language were
FORTRAN, unless you program in assembler. There is also languages like
Pascal, Basic, Lisp, Snobol, Focal and some other stuff.
Basically, PDP-8 were often in labs, but also saw a lot of use as
control system for everything from print shops to running billboards.

The PDP-11 was designed as a replacement for the PDP-8. It's a 16-bit
machine. Still very easy to program, but way more advanced than the
PDP-8. It has a very easy to use instruction set which is very CISC.
Lots of registers (for the time), and a higher density for the code, in
addition to the ability to address much more memory. Maximum for the
architecture is 4Mbyte. The PDP-11 was both popular as a lab machine in
the same style of environment as the PDP-8, but also very commonly seen
as a timesharing system. Unix more or less originated on the PDP-11, and
the architecture was the most popular minicomputer of the 70s. It has a
wide range of different models with different price/performance point.
Many different operating systems, some for single user, others for up to
roughly 100 concurrent users. FORTRAN was a very popular choice in the
early days, but you also have Basic, Cobol, Forth, C, Pascal, Lisp,
Prolog, Focal, Dibol, as well as database applications such as
Datatrieve or Oracle, and lots of communication possibilities. It was
for quite some time a workhorse in all kind of places, and you might
still find some out there running things.

The PDP-10 was designed in the early 60s as a large timesharing machine.
As such, it's capabilities were designed for a different workload. You
commonly had front-end computers that offload the main CPU from a lot of
I/O work, and the possibility of a lot more memory. It's a very
different concept than a PDP-11, and you'd basically never see a PDP-10
as a one user system. It became very popular in academia, and a lot of
AI and CS research happened on PDP-10s. There were a few different OSes
available, and as far as languages go, you probably could find almost
any language you could think of.

Johnny

On 2025-03-20 23:13, DR wrote:
> You are getting to the heart of my question, then.
>
>
> My experience with computers of that era were the CDC 3600 at the U
> which rain mainly physics jobs.  Word length and architecture seemed to
> make that a choice for those needing computer resources.
>
>
> The other machine was a Univac 1108, which was general purpose to some
> degree as I  understood it.  The academic branch seemed to use it to
> teach languages, programming techniques, etc.  I guess that scientific
> purposes would be served, but the majority of output of that type were
> tapes created to drive a CalComp plotter and do print jobs.  I don't
> recall starting jobs from the physicists, at least until the CDC3600 got
> blown  up by the four anti-war guys.
>
>
> The B5500 was in the corner, almost never went down, and the operator on
> that console was sure to bring lots of reading materials or classwork to
> occupy themselves.  I think it was oriented towards business needs, but
> it was hard to know.  I'm not even sure what types of jobs or languages
> it supported.  All I recall is hearing that it was very good at what it
> did.  We did not load academic (language training, etc.) onto thatmachine.
>
>
> There were a few IBM 360s scattered around, but not sure if they did
> anything other that statistics in some  of the social sciences areas in
> their buildings.
>
>
> The PDP-8 was brought in to have a more hands-on machine.  Most ofthe

> time we got  to play with FOCAL and other things  on it, morein support

> of language learning.
>
>
> So the question then was a general orientation that DEC designed each
> machine for?  General computing? More scientific analysis? Long word
> length to get precision out of each instruction?
>
>
> Since then there seems to be the concept that a computer is more general
> purpose, or at least what I perceive.
>
>
> And then back to the original question, if I had a 'stock' -10 and a
> stock -11 sitting on the floor, and had a FORTRAN program, could both
> run it, and  if so would they run with equal speed or alacrity forthe
> mission?  Would one be optimized to run LISP or SNOBOL or PASCAL or
> FORTRAN, etc?  COBOL (that's one that the B5 seemed to run, along with a
> few of the 360s in other parts of the campus.)
>

[Peter Whisker]

The PDP-10 was a mainframe rather thana minicomputer, the kind of thing you ran a University Computer laboratory on. At Trinity College Dublin where I studied, they retired their old 1960's IBM S/360-44 in 1978 and replaced it by a DECSYSTEM/20 (PDP-10) which gave timeshare access to dozens of students at a time using VT-52 terminals. I don't think a PDP-11/70 would necessarily have fulfilled the same role as successfully.

Peter

--
You received this message because you are subscribed to the Google Groups "[PiDP-11]" group.

To unsubscribe from this group and stop receiving emails from it, send an email to pidp-11+u...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/pidp-11/bd0cf7d3-f611-4f93-a012-ddbfda9a1d17%40gmail.com.

[Peter Whisker]

https://www.tcd.ie/itservices/about/50th-anniversary/#tab_1735958

[Clem Cole]

On Thu, Mar 20, 2025 at 5:08 PM Charles Ess <charl...@gmail.com> wrote:

I'm also guessing that no one has decided they need to code the emulator of the FP11-F that could plug into the PDP11 as simulated in simh and run on the Pi ...

https://www.tuhs.org/cgi-bin/utree.pl?file=V7/usr/src/libfpsim

The FP-11 simulator for UNIX has been running fine since the early/mid-70s when it was first created.  I think that is the version from early 1979 for dates. It will run under simh on a pi just fine.

If you are interested in FP, I >>highly<< recommend looking at the source of Jack Burness's Moonlander.  That 11/05 lacked an FP, so he has a complete set of 16-bit integer transcendental using the cordic method so he completely avoided using it.  http://www.brouhaha.com/~eric/retrocomputing/dec/gt40/

ᐧ

[DR]

Beautiful summary, bringing forward facts that may be intuitive to many
here, but this discussion was refreshing and helpful in differentiating
what these famous boxes did and why they were chosen for certain tasks.


Thank you.


Fun, ain't it?

[timr...@gmail.com]

Depends on what you mean by stock 10, or stock 11.  A very basic 11 could mean paper tape only.  No disk, and limited core.  I doubt most places bought just the lowest end machine.  Probably would be good to clarify.  How much memory?  Disks?  Tape only?  Etc.  For both.  The original 10's were core only.  Some of the early 11s were too. But I am sure later on 11s had fully semiconductor memory.  Core was very expensive.  Disks were too.  Yes, both 10s and 11s had Fortran support.  Burroughs machines were usually optimized for math type stuff so banks love them.

[terry-...@glaver.org]

On Thursday, March 20, 2025 at 7:41:48 PM UTC-4 cl...@ccc.com wrote:

On Thu, Mar 20, 2025 at 5:08 PM Charles Ess <charl...@gmail.com> wrote:

I'm also guessing that no one has decided they need to code the emulator of the FP11-F that could plug into the PDP11 as simulated in simh and run on the Pi ...

https://www.tuhs.org/cgi-bin/utree.pl?file=V7/usr/src/libfpsim

The FP-11 simulator for UNIX has been running fine since the early/mid-70s when it was first created.  I think that is the version from early 1979 for dates. It will run under simh on a pi just fine.

True. But since simh can emulate a PDP-11 with the appropriate FP processor, the only need for emulation of FP in the Unix kernel was when running on real (or emulated) CPU models that lacked hardware FP. If you had a model where FP was optional, in simh you just told it you'd "bought" the multi-thousand-dollar FP add-on. Emulation in simh is likely substantially faster than executing Macro-11 FP subroutines, since you're eliminating one level of emulation.

And you can really confuse many of the PDP-11 operating systems if you modify simh to allow configurations that were never possible on production DEC hardware. RSTS/E will "fall down, go *SPLAT*" if it encounters an 11/70 with CIS, for example. I expect Johnny's RSX-11M+ will behave better, since M+ got to play with all of the cool prototype hardware (11/70MP, KE74, etc.).

BTW, the 11/23 and 11/24 (same chipset) had both the KEF11 (which implemented FP in microcode and used scratchpad registers in the MMU) and the FPF11 which was another whole board with a gaggle (16) of AM2901 bit-slice processors and dedicated scratchpad registers on-board. Trivia - the FPF11 (M8188) was likely the only DEC card that could be used in both Unibus and Q-bus systems (all it got from the bus was DC power, and that was jumperable based on bus type).

The closest DEC came in later 11's was things like the DCJ11 CPU which had microcoded floating point and could optionally be equipped with a FPJ11 accelerator (down to one chip due to being a lot newer).ᐧ

[Milo Velimirović]

On Mar 21, 2025, at 3:06 PM, pid...@googlegroups.com wrote:

And no, the MMU does not affect performance.

Not completely correct; it would slow every mapped memory reference. Enabling the KT11 on early pdp11s (11/40, 11/45, and probably a few others that lacked a cache,) would add time to every mapped memory reference. This time was on the order of 100nS; in the era of 1µS memory this was considered an acceptable performance hit. This time was basically the time it took for the KT11 to generate a physical address when presented with a virtual address.

—Milo

[Johnny Billquist]

Sorry, but no. The fact that you have additional gate delays are not
relevant. The system runs at the system clock, and the number of cycles
are not affected if MMU is enabled or not, so the code will execute at
the same speed no matter if the MMU is on or off. The delays are not
interesting, because the speed of the machine is determined by the
clock, and not gate delays.

So performance is not affected by the MMU being on or off. Run your
program in both scenarios, and it will take the same amount of time.

Johnny

[Johnny Billquist]

If people are really interested in the performance topic, the PDP-11
processor handbooks have excellent breakdowns of the time to execute all
kind of instructions, so you can actually calculate the amount of time
needed for any instruction you have. And MMU on/off is not in the equations.

Johnny

[Johnny Billquist]

And to quote the PDP-11/70 processor handbook from 1977-1978:

C.1.4 NOTES
1. The times specified generally apply to Word instructions. In most
cases Even Byte instructions have the same time, with some Odd
Byte instructions taking longer. All exceptions are noted.
2. Timing is given without regard for NRP or BR serving. Core memory
is assumed to be iocated within the first 128K memory unit.
3. Times are not affected if Memory Management is enabled.
4. All times are in microseconds.


Specifically note point 3.

Johnny

[Johnny Billquist]

And just to throw some more data into the thread.

On the 11/40, timing is affected if an MMU is *installed* or not. Don't
matter if it's enabled or not.
So there you have a delay introduced by all signals having to go through
more logic in general.

But still no performance hit because of mapped or unmapped references as
such.

Johnny

[Johnny Billquist]

Ah. One final one, and then I'll be quiet.

On the 11/45, you appear to be correct.

There it's stated that if the KT-11C is installed and operating, you
have a instruction time increase of .09us for each memory reference.

Didn't know that some models actually did have a performance hit. Oh
well, I learned something new today. Thanks. :-)

Johnny