http://story.news.yahoo.com/news?tmp...21/tc_nf/33261

IBM's also doing multicore chips, in the Power processor family. The
current version's mostly being aimed at the gaming market and can't
easily be turned into a multi-core Macintosh, but it might be
interesting to see what the linux hackers could do with it.

Joe Kesselman wrote:
IBM's also doing multicore chips, in the Power processor family. The
current version's mostly being aimed at the gaming market and can't
easily be turned into a multi-core Macintosh, but it might be
interesting to see what the linux hackers could do with it.


So is this more-or-less the same, performance-wise as a dual processor
system? It sounds basically like two chips in one socket.

Jonny Durango

Jonny Durango wrote:
Joe Kesselman wrote:
IBM's also doing multicore chips, in the Power processor family. The
current version's mostly being aimed at the gaming market and can't
easily be turned into a multi-core Macintosh, but it might be
interesting to see what the linux hackers could do with it.

So is this more-or-less the same, performance-wise as a dual processor
system? It sounds basically like two chips in one socket.

It _could_ be a little faster because of the reduced propagation delay.
But, it _should_ be a lot cheaper, since you're paying for only one piece
of silicon and one socket.
--scott

--
"C'est un Nagra. C'est suisse, et tres, tres precis."

Scott Dorsey wrote:
Jonny Durango wrote:

Joe Kesselman wrote:

IBM's also doing multicore chips, in the Power processor family. The
current version's mostly being aimed at the gaming market and can't
easily be turned into a multi-core Macintosh, but it might be
interesting to see what the linux hackers could do with it.

So is this more-or-less the same, performance-wise as a dual processor
system? It sounds basically like two chips in one socket.


It _could_ be a little faster because of the reduced propagation delay.
But, it _should_ be a lot cheaper, since you're paying for only one piece
of silicon and one socket.
--scott


This of course begs the question, will MB manufacturers start making
dual, dual proc boards? *drool*

Jonny Durango

Scott Dorsey wrote:
It _could_ be a little faster because of the reduced propagation delay.

There are also questions of exactly how the processor cores are able to
talk to each other. Given that the company has been making world-leading
massively multiprocessing systems I suspect there's likely to be some
carry-over of concepts into these smaller systems... maybe not now, but
in the future.

(Claimer: I'm an IBMer so I'm somewhat biased. Disclaimer: I know none
of the details of these particular chips beyond what I've seen in the
press, and it's been over 15 years since I was mucking about with
supercomputer circuit designs so I'm woefully out of date. Do NOT assume
I've got more of a clue than anyone else.)

Joe Kesselman wrote:

(Claimer: I'm an IBMer so I'm somewhat biased. Disclaimer: I know none
of the details of these particular chips beyond what I've seen in the
press, and it's been over 15 years since I was mucking about with
supercomputer circuit designs so I'm woefully out of date. Do NOT assume
I've got more of a clue than anyone else.)

I thought the 360/91 was the last time IBM was mucking around with
supercomputer circuit designs....
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."

Scott Dorsey wrote:
I thought the 360/91 was the last time IBM was mucking around with
supercomputer circuit designs....

You haven't been looking at what's happening recently with Blue Gene and
its kin. Not traditional superscalar or vector architecture, but it
cranks out one honking huge number of MIPS and they've figured out how
to distribute tasks efficiently across it.

Scott Dorsey wrote:
Joe Kesselman wrote:

(Claimer: I'm an IBMer so I'm somewhat biased. Disclaimer: I know
none of the details of these particular chips beyond what I've seen
in the press, and it's been over 15 years since I was mucking about
with supercomputer circuit designs so I'm woefully out of date. Do
NOT assume I've got more of a clue than anyone else.)

I thought the 360/91 was the last time IBM was mucking around with
supercomputer circuit designs....

Things have changed so much since the 360/91-2-3 that we wouldn't even
recognize it as being a supercomputer today. Of course circuitry is a
lot faster, but even the cheapest modern PC has more sophistication
when it comes to pipelineing, cacheing, and parallelism. Shortly,
multiprocessing will be added to that list.

Arny Krueger wrote:

Things have changed so much since the 360/91-2-3 that we wouldn't even
recognize it as being a supercomputer today. Of course circuitry is a
lot faster, but even the cheapest modern PC has more sophistication
when it comes to pipelineing, cacheing, and parallelism. Shortly,
multiprocessing will be added to that list.

Yet all of those features (and more) except multiprocessing
were pioneered with the 91 (or was the first one the 90?)


Bob
--

"Things should be described as simply as possible, but no
simpler."

A. Einstein

Bob Cain wrote:
Arny Krueger wrote:

Things have changed so much since the 360/91-2-3 that we
wouldn't
even recognize it as being a supercomputer today. Of
course
circuitry is a lot faster, but even the cheapest modern
PC has more
sophistication when it comes to pipelineing, cacheing,
and
parallelism. Shortly, multiprocessing will be added to
that list.

Yet all of those features (and more) except
multiprocessing
were pioneered with the 91 (or was the first one the 90?)

My impression is that the 360/67 was IBM's first production
multiprocessor.

I worked for IBM in those days, and heard rumors of at least
one earlier computer complex that joined two 709x into a
multiprocessor.

Our field office SE doc said that central corporate approval
was required to sell a 360/67 to a client because of its
special technical support needs on the part of the customer
and IBM. This was a nice way of saying that they really
didn't have a working OS for it at the time.

Heck in those days, OS/360 wouldn't run long enough to
generate itself.

I believe that I may have earlier unknowingly actually seen
such a 7094 complex at the GM Engineering complex in Warren
Michigan, when I worked there.

The 360/90 is credited with being the first computer that
supported pipelining with out-of-order execution.

http://citeseer.ist.psu.edu/context/59581/0

"The concept of out of order execution was first implemented
in the IBM 360 90 [Ande67]"

Arny Krueger wrote:

Yet all of those features (and more) except

multiprocessing

were pioneered with the 91 (or was the first one the 90?)


My impression is that the 360/67 was IBM's first production
multiprocessor.

That sounds correct. I should remember for sure since it
was built in the same lab space where I was part of the
370/155 and 158 design team, but I don't.

It just dropped into my memory that the first prototype was
labled the 360/44 and was based on the 360/40, a
particularly simple version of a System 360.

I worked for IBM in those days,

Me too. I was in SDD starting in Poughkeepsie in '67.
Where and when was your service?

and heard rumors of at least
one earlier computer complex that joined two 709x into a
multiprocessor.

Never heard of that. Could be, though.

Our field office SE doc said that central corporate approval
was required to sell a 360/67 to a client because of its
special technical support needs on the part of the customer
and IBM. This was a nice way of saying that they really
didn't have a working OS for it at the time.

You got that right. I didn't think they ever went
commercial with it, that it was just a feasability project
but there is some dropout in my memory of things that
happened 30+ years ago. :-)

Heck in those days, OS/360 wouldn't run long enough to
generate itself.

I don't remember that either and I used it a lot in my
design work. Seems to me that by '67, at least, it was rock
solid in terms of crashing but there were a lot of
functional bugs.

I believe that I may have earlier unknowingly actually seen
such a 7094 complex at the GM Engineering complex in Warren
Michigan, when I worked there.

The 360/90 is credited with being the first computer that
supported pipelining with out-of-order execution.

And virtual registers and super-scaler architecture and
branch prediction, and... It was a long time before those
features re-emerged in other products, notably the Intel
Pentium line and IBM's Power PC architecture (who's
architect, John Cocke, was a personal friend and pub buddy.)

Thanks for the reminiscence. :-)


Bob
--

"Things should be described as simply as possible, but no
simpler."

A. Einstein

Bob Cain wrote:
Arny Krueger wrote:

Yet all of those features (and more) except

multiprocessing

were pioneered with the 91 (or was the first one the
90?)

My impression is that the 360/67 was IBM's first
production
multiprocessor.

That sounds correct. I should remember for sure since it
was built in the same lab space where I was part of the
370/155 and 158 design team, but I don't.

It just dropped into my memory that the first prototype
was
labled the 360/44 and was based on the 360/40, a
particularly simple version of a System 360.

Interesting. I would have thought that the 67 was based on
the 65.

I worked for IBM in those days,

Me too. I was in SDD starting in Poughkeepsie in '67.
Where and when was your service?

Flint, MI field office, about 9 months in 66, before I was
drafted.

and heard rumors of at least
one earlier computer complex that joined two 709x into a
multiprocessor.

Never heard of that. Could be, though.

Our field office SE doc said that central corporate
approval
was required to sell a 360/67 to a client because of its
special technical support needs on the part of the
customer
and IBM. This was a nice way of saying that they really
didn't have a working OS for it at the time.

You got that right. I didn't think they ever went
commercial with it, that it was just a feasability project
but there is some dropout in my memory of things that
happened 30+ years ago. :-)

I know of a number of non-IBM shops that had 67s. GM &
University of Michigan used them to run TSS & MTS
respectively. I seem to recall that Princeton and Cornell
had them as well. They ran CP-67 as I recall.

Heck in those days, OS/360 wouldn't run long enough to
generate itself.

I don't remember that either and I used it a lot in my
design work. Seems to me that by '67, at least, it was
rock
solid in terms of crashing but there were a lot of
functional bugs.

Yes, but that was one year later than 1966.

I believe that I may have earlier unknowingly actually
seen
such a 7094 complex at the GM Engineering complex in
Warren
Michigan, when I worked there.

The 360/90 is credited with being the first computer that
supported pipelining with out-of-order execution.

And virtual registers and super-scaler architecture and
branch prediction, and... It was a long time before those
features re-emerged in other products, notably the Intel
Pentium line and IBM's Power PC architecture (who's
architect, John Cocke, was a personal friend and pub
buddy.)

Thanks for the reminiscence. :-)

Here's mo

http://www.beagle-ears.com/lars/engi...t/model360.htm




Bob

"Arny Krueger" wrote in news:N9Cdndfd2b7NPeXfRVn-
:

Bob Cain wrote:
Arny Krueger wrote:

My impression is that the 360/67 was IBM's first
production
multiprocessor.

That sounds correct. I should remember for sure since it
was built in the same lab space where I was part of the
370/155 and 158 design team, but I don't.

It just dropped into my memory that the first prototype
was
labled the 360/44 and was based on the 360/40, a
particularly simple version of a System 360.

Interesting. I would have thought that the 67 was based on
the 65.


Could it be that your are confounding the multiple apparent machines made
possible by VM with physical multiprocessing in the case of the '67?

My recollection is that the "big new thing" on the 360/67 was virtual
memory. I actually ran a little APL on the campus one from a Selectric
terminal somewhere in building 10 around 1970, but don't know where the
system lived. Mean time to crash was something like a couple of hours at
the time.

Here is a current reference which claims it was based on a 65--I assumed it
was at the time, but had no direct knowledge.

http://www.multicians.org/thvv/360-67.html

Peter Stoll

Peter A. Stoll wrote:
"Arny Krueger" wrote in
news:N9Cdndfd2b7NPeXfRVn-
:

Bob Cain wrote:
Arny Krueger wrote:

My impression is that the 360/67 was IBM's first
production
multiprocessor.

That sounds correct. I should remember for sure since
it
was built in the same lab space where I was part of the
370/155 and 158 design team, but I don't.

It just dropped into my memory that the first prototype
was
labled the 360/44 and was based on the 360/40, a
particularly simple version of a System 360.

Both the 360/30 and 360/40 were very microcode-centric
machines. They almost had no native hardware fun ctions. You
could rewrite the microcode (using a keypunch for the
360/30) and it would be just about any byte-oriented machine
you wanted it to be. I believe that some customers actually
did that.

Interesting. I would have thought that the 67 was based
on
the 65.

Could it be that your are confounding the multiple
apparent machines
made possible by VM with physical multiprocessing in the
case of the
'67?

No. I would need to be far older to make that mistake. ;-)

My recollection is that the "big new thing" on the 360/67
was virtual
memory.

I understand that inside IBM's labs, their earlier prototype
VM machines were modified 360/40s. Before that there were VM
simulations run on 7094s.

I actually ran a little APL on the campus one from a
Selectric terminal somewhere in building 10 around 1970,
but don't
know where the system lived. Mean time to crash was
something like a
couple of hours at the time.

In 1970, this might have been a 370/145 or some such. As I
recall it had a lot of APL implemented in microcode.

Here is a current reference which claims it was based on a
65--I
assumed it was at the time, but had no direct knowledge.

http://www.multicians.org/thvv/360-67.html

Thanks. Highly enlinghtening.

You might find it interesting that Melinda Varian mentioned
there was an associate of mine, and mentioned me in a
well-known paper about OS software maintenance that she
published within Share.

http://pucc.princeton.edu/~melinda/tutorial.pdf

Try the following retrieval to see the 360/67 -
multiprocessor connection:

http://www.google.com/search?hl=en&q...multiprocessor