On Oct 30, 12:49 pm, eug... at (no spam) cse.ucsc.edu (Eugene Miya) wrote:

The people who thought highly of the CM-1 were the AI types who also
liked Symbolics 3600s.  I also saw and have pieces of the MPP (now at
the Smithsonian near Dulles). The AI types remain largely clueless to
parallelism.  They don't constitute a viable market.

The real ho-hum part wasn't the speed. The problem for real machines
with real problems was that they lacked sufficient amounts of real memory.

That's an interesting claim. It implies that all you have to do is to
build a knock-off of CM-1, give it sufficient memory, and people would
no longer regard Thinking Machines as the exercise in hubris that they
now do. I think of Thinking Machines as a major contributor to moving
computing out of Kendall Square forever.

I read Danny Hillis' PhD thesis after conversations with you. I don't
remember much of it because I'm pretty sure I missed the really,
really smart part of it. Mostly, it seemed really, really obvious.
Perhaps I'm just not really, really smart enough myself.

In any case, you have offered one of your as-always tantalizing
claims. Is there any substantiating evidence, or are you just
repeating what the really, really smart people were saying at the
time?

Probably, I'll do better just to go to AI seminars at MIT and see what
I can find out for myself. I'm sure many of the players are still
around.

Robert.

Joe Pfeiffer wrote:
"Andy \"Krazy\" Glew" <ag-news at (no spam) patten-glew.net> writes:
3) I think I remember the UW Pyramid; Pardo?

It was a design Steve Tanimoto at the University of Washington was
working on with his students (of which I was one) in the early 1980s.
The idea was that instead of a grid topology, the machine would have a
series of grids, each 1/4 the size of the one below it: so if the base
was 128x128, the next was 64x64, the next 32x32, and so forth. In
addition to being connected to its eight neighbors in the plane, a node
would also be connected to its parent and its four children.

What's the era for that? It sounds very much like what DARPA tried to
turn David Shaw's tree machine into just before they cut his
funding. (Everyone, including DARPA, knew that the interconnect
wouldn't scale, but apparently they insisted on the change anyway...)

Robert Myers wrote:

A criticism you could make of most of the clever "scalable" machines
that have been built.  Well, you can do lots and lots of low order
finite differences or finite elements, for which purpose a hypercube
is overkill.

I've spent a bit of time thinking about how to do the really hard
stuff of moving data around in big machines, which is one reason why I
am so touchy about all the non-breakthroughs that are mostly press
releases.

At this point, the approach that seems most promising to me is exactly
the opposite of Blue Gene.  Pack as much muscle and routing onto
single chips and single boards as you can, so that you need as little
of the expensive global interconnect as possible, and so that you can
afford global interconnect that isn't a joke.  From that perspective,
GPGPU's look like a really attractive computing engine.

Robert.

If you don't like blue gene, you can be accommodated.

-----------------------
The world’s fastest supercomputer at Los Alamos National Laboratory is
more than double the speed and scale of the previous benchmark by
breaking through the "petaflop barrier" of 1,000 trillion operations per
second. That's nearly three times faster than the leading contenders on
the November 2007 TOP500 list of supercomputers worldwide. Yet compared
to most traditional supercomputers of today, hybrid design of the
supercomputer at Los Alamos delivers world-leading energy efficiency, as
measured in flops per watt.

Designed by IBM, the world's first "hybrid" supercomputer introduces the
use of the IBM PowerXCell™ 8i chip, an enhanced Cell Broadband Engine™
(Cell/B.E.™) chip—originally developed for video game platforms—in
conjunction with x86 processors from AMD™. The supercomputer was built
for the U.S. Department of Energy's National Nuclear Security
Administration and will be housed at Los Alamos National Laboratory in
New Mexico.

Inside the supercomputer

In total, the supercomputer at Los Alamos connects 6,948 dual-core AMD
Opteron™ chips and 12,960 PowerXCell 8i processors. The Opteron
processors handle standard processing such as file system I/O. The
PowerXCell 8i processors accelerate mathematical and CPU-intensive
processing.

Two PowerXCell 8i-based blade servers (IBM BladeCenter® QS22) and one
AMD-based blade (IBM BladeCenter LS21) are integrated into a specialized
"tri-blade" configuration. The machine is composed of a total of 3,456
tri-blade units, each of which can run at 400 billion operations per
second (400 gigaflops).

Its 10,000 connections—both InfiniBand® and Gigabit Ethernet—require 57
miles of fiber optic cable. The system has 80 terabytes of memory,
weighs 500,000 pounds, and is housed in 288 refrigerator-sized, IBM
BladeCenter racks occupying 6,000 square feet.
----------------------------------

(not sure what this one is made out of.  May find out at celebration for
national medal of technology wednesday)

IBM computer will have power of 2 million laptops

SAN FRANCISCO (Reuters) - Seven months after IBM delivered the world's
fastest supercomputer, it has announced an even speedier one with the
computing power of 2 million laptops.

IBM said on Tuesday it is developing the technology for its new Sequoia
computer, with delivery scheduled in 2011 to the Department of Energy
for use at Lawrence Livermore National Laboratory.

Sequoia will chug along at 20 petaflops per second and is one order of
magnitude quicker than its predecessor. The earlier machine, delivered
in June to the Energy Department, broke the 1 petaflop barrier.

Peta is a term for quadrillion and FLOP stands for floating point
operations per second.

Sequoia, and a smaller computer called Dawn, are being built in
Rochester, Minnesota, for use in simulating nuclear tests. IBM says they
can also be used for complex tasks like weather forecasting or oil
exploration.

IBM says Sequoia will be highly energy-efficient for the job it does but
even so will occupy 96 refrigerator-sized racks in an area the size of a
big house -- 3,422 square feet (318 square meters).
(from another article)
While Dawn is based on currently available processors, Sequoia will be
built on future IBM BlueGene technology.

When completed, the supercomputer will have 1.6 Petabytes of memory, 96
racks, 98,304 compute nodes, and 1.6 million cores. IBM promises Sequoia
will be 160 times more power efficient than ASC Purple and 17 times more
so than BlueGene/L - both previously installed supercomputers at LLNL.

(program that you whiners)  
--------------------------------------------------

The National Oceanic and Atmospheric Administration (NOAA) has completed
a nine-year, $180 million IBM supercomputer build-out for weather and
climate prediction.

The two new systems, dubbed Stratus and Cirrus, will let NOAA run more
complex models in an effort tol boost the nation's watch and warning
lead times for tornadoes, floods, hurricanes, tsunamis, and winter storms..

The two supercomputers run as primary and backup in tandem, and are both
based on IBM Power 575 Systems. The main Stratus supercomputer will
process gigabytes of weather data each day, including temperature, wind,
atmospheric pressure, and satellite and oceanographic data.

The computers are capable of making 69.7 trillion calculations per
second. NOAA said that's the equivalent of one person with a calculator
working for three million years.
-------------------------------------------

Someone wants something, my buddies on the tundra will build them one.  :-)

del

On Nov 2, 11:52 am, Joe Pfeiffer <pfeif... at (no spam) cs.nmsu.edu> wrote:


Everybody knew they were being propped up by DARPA; that was no secret
at all. Would anybody have ever made any money on supercomputers
without lots of government money?

I don't see how that made it an exercise in hubris, though.


The NSA has at times assured Cray that there would a market for
certain capabilities.

That's different from spec'ing a machine from your office in
Arlington, VA and telling the other customers you fund that that's
what they need to buy with your money.

From my end of the telescope, the story is always pretty much the
same: someone who knows lots about electronics and very little about
computation builds another strange beast with lots of useless gates
and then shrugs at the waste of taxpayer money. I'm repeating myself
here, but Thinking Machines was no more egregious than some of LLNL's
buying sprees at IBM.

The MIT AI laboratory is one of the few places I know of where people
(at least at one time) knew a bit about both hardware and a certain
kind of computation--albeit one that I know little about. Maybe there
is something there, but I haven't gotten it yet.

I only know bits and pieces of the Connection Machine story. Mostly
it seemed like trading on connections and media hustle, rather than
anything to do with real technical insight. There were big names
involved, and I persist in thinking that that and the glamor factor
were more important than any real technical insight or achievement.

snip

To see an SIMD machine whose topology was a hypercube, with actual
routing possible, seemed just amazing.


Maybe I just understand too little about routing algorithms to
understand what's amazing about that.


Unfortunately, it turned out to
be too hard to actually *do* anything with that capability....


A criticism you could make of most of the clever "scalable" machines
that have been built. Well, you can do lots and lots of low order
finite differences or finite elements, for which purpose a hypercube
is overkill.

I've spent a bit of time thinking about how to do the really hard
stuff of moving data around in big machines, which is one reason why I
am so touchy about all the non-breakthroughs that are mostly press
releases.

At this point, the approach that seems most promising to me is exactly
the opposite of Blue Gene. Pack as much muscle and routing onto
single chips and single boards as you can, so that you need as little
of the expensive global interconnect as possible, and so that you can
afford global interconnect that isn't a joke. From that perspective,
GPGPU's look like a really attractive computing engine.

Robert.

Someone wants something, my buddies on the tundra will build them one. :-)

del

What's the era for that?

Del Cecchi` wrote:
Someone wants something, my buddies on the tundra will build them
one. :-)

del

Tundra? Is Blue Gene built in Minnesota?

I'm Baaaaak.   Turns out the software was like 30 or 40 guys in
Minnesota and like 15 in Yorktown.

Just to be clear, as I understand it, the architecture and main idea for
  blue gene came from the research division.  Much of the implementation
and productization happened in Rochester.  Quite a number of other sites
were also involved in one way or another.

you all might find this of interest if bombs aren't your fieldhttp://bluebrain.epfl.ch/

Robert Myers wrote:
On Nov 4, 7:33 pm, Del Cecchi <delcecchinospamoftheno... at (no spam) gmail.com
wrote:

I'm Baaaaak.   Turns out the software was like 30 or 40 guys in
Minnesota and like 15 in Yorktown.

Just to be clear, as I understand it, the architecture and main idea for
  blue gene came from the research division.  Much of the implementation
and productization happened in Rochester.  Quite a number of other sites
were also involved in one way or another.

you all might find this of interest if bombs aren't your fieldhttp://bluebrain.epfl.ch/

I'm sure that LLNL would like to move brain research behind the barbed-
wire fence, too, and it would not surprise me at all if they hadn't
taken a shot at it, and I expect that the move will come sooner or
later if it hasn't already.

We've talked about Blue Brain here before, and it pulled a couple of
interesting lurkers out of the woodwork.  Yes, you can use "super"-
computers for all kinds of things besides building bombs (or combat
aircraft, or lasers that burn holes in things or set off thermonuclear
explosions).  What might be the best architecture is, at this moment,
a more wide open question than at any other time in recent memory, in
my ever-humble opinion.

I'm glad to see that IBM is in there pitching.

Robert.

Sure.  We do hetrogenous. We do power efficient small cores in large
numbers.  We do high frequency big cores.  Probably some other stuff
too.  ( we used rather loosely although I still feel a kinship with my
former co-workers)

I would guess that, if there is any approach to high performance
computing mentioned, someone at IBM has probably taken a good look at
it.  They may not have made the correct decision, but they thought about
it.

And something about the computing with GPU thing that just crossed my
mind.  I have decided that it is a kluge for those without the resources
to design a chip of their own.  After all the existing GPU designs are
not optimized for use in large scale computing.  However something much
like them certainly could be, (before I am attacked).

In the meantime they are busy on Sequoia and Dawn and who knows what else..

Del Cecchi` wrote:
Someone wants something, my buddies on the tundra will build them
one. :-)

del

Tundra? Is Blue Gene built in Minnesota?

On Nov 4, 7:33 pm, Del Cecchi <delcecchinospamoftheno... at (no spam) gmail.com
wrote:

I'm Baaaaak. Turns out the software was like 30 or 40 guys in
Minnesota and like 15 in Yorktown.

Just to be clear, as I understand it, the architecture and main idea for
blue gene came from the research division. Much of the implementation
and productization happened in Rochester. Quite a number of other sites
were also involved in one way or another.

you all might find this of interest if bombs aren't your fieldhttp://bluebrain.epfl.ch/

I'm sure that LLNL would like to move brain research behind the barbed-
wire fence, too, and it would not surprise me at all if they hadn't
taken a shot at it, and I expect that the move will come sooner or
later if it hasn't already.

We've talked about Blue Brain here before, and it pulled a couple of
interesting lurkers out of the woodwork. Yes, you can use "super"-
computers for all kinds of things besides building bombs (or combat
aircraft, or lasers that burn holes in things or set off thermonuclear
explosions). What might be the best architecture is, at this moment,
a more wide open question than at any other time in recent memory, in
my ever-humble opinion.

I'm glad to see that IBM is in there pitching.

Robert.

And something about the computing with GPU thing that just crossed my
mind. I have decided that it is a kluge for those without the resources
to design a chip of their own. After all the existing GPU designs are
not optimized for use in large scale computing. However something much
like them certainly could be, (before I am attacked).

Robert Myers wrote:

IBM did eventually develop a specialized processor for Cray

Are you sure? For Cray? Wasn't that DEC?

The only thing I can think of that IBM did for an outside supercomputer
was what they did for Steve Chen/SSI.

So Tesla is used as a bag on the side of a convention system?

i don't remember company ever spending any money on cray