Blue Gene/P
Del Cecchi
IBM Triples Performance of World's Fastest, Most Energy-Efficient
Supercomputer
Tuesday June 26, 8:00 am ET
New Blue Gene/P Designed to Break the "Quadrillion" Barrier
ARMONK, NY--(MARKET WIRE)--Jun 26, 2007 -- IBM (NYSE:IBM - News) today
announced Blue Gene/P, the second generation of the world's most
powerful supercomputer. Blue Gene/P nearly triples the performance of
its predecessor, Blue Gene/L -- currently the world's fastest computer
-- while remaining the most energy-efficient and space-saving computing
package ever built.
(snip)
The IBM® System Blue Gene®/P Solution scales to operate continuously at
speeds exceeding one "petaflop" -- or one-quadrillion operations per
second. (snip) The Blue Gene/P supercomputer can be configured to reach
speeds in excess of three petaflops, a performance level that many
thought unattainable only a few short years ago.
(snip)
A Green Design Ahead of its Time
The Blue Gene® supercomputer line was born from a visionary IBM
initiative to develop a hugely scalable and highly reliable scientific
computing platform. With Blue Gene, designers sidestepped two key
constraints on state-of-the-art supercomputing -- power usage and space
requirements. The Blue Gene supercomputer was purpose-built to fit in
smaller spaces and use less electricity compared to other commercially
available designs. Today, the Blue Gene/P supercomputer is at least
seven times more energy efficient than any other supercomputer.
The influence of the Blue Gene supercomputer's energy-efficient design
and computing model -- once considered exotic -- can be seen everywhere
in the industry where people have attempted to lower energy use and get
performance without traditional reliance on chip frequency. The
breakthrough BlueGene supercomputer design uses many small, low-power
embedded chips each connected through five specialized networks inside
the system.
Some of the world's leading research laboratories and universities have
already placed orders for Blue Gene/P supercomputers. The U.S. Dept. of
Energy's Argonne National Laboratory, Argonne, Ill., will deploy the
first Blue Gene/P supercomputer in the U.S. beginning later this year.
In Germany, the Max Planck Society and Forschungszentrum Julich also
plan to begin installing Blue Gene/P systems in late 2007. Additional
Blue Gene/P system rollouts are being planned by Stony Brook University
and Brookhaven National Laboratory in Upton, N.Y., and the Science and
Technology Facilities Council, Daresbury Laboratory in Cheshire, England.
(snip)
At FZ Julich, where researchers have been using a Blue Gene/L machine
for two years, a Blue Gene/P system will allow for more breakthrough
science -- in such areas as particle physics and nanotech, for example
-- while keeping the research facility within acceptable power budgets.
"The big computing power at low electricity rates allows us to boost the
performance of very complex and computationally intensive algorithms,"
said Thomas Lippert, director of the supercomputing center at FZ Julich.
Inside the Fastest Computer Ever Built
Like its predecessor, the Blue Gene/P supercomputer is a modular design,
composed of "racks" that can be added as requirements grow.
Four IBM® (850 MHz) PowerPC® 450 processors are integrated on a single
Blue Gene/P chip. Each chip is capable of 13.6 billion operations per
second. A two-foot-by-two-foot board containing 32 of these chips churns
out 435 billion operations every second, making it more powerful than a
typical, 40-node cluster based on two-core commodity processors.
Thirty-two of the compact boards comprise the 6-foot-high racks. Each
rack runs at 13.9 trillion operations per second, 1,300 times faster
than today's fastest home PC.
The one-petaflop Blue Gene/P supercomputer configuration is a
294,912-processor, 72-rack system harnessed to a high-speed, optical
network. The Blue Gene/P system can be scaled to an 884,736-processor,
216-rack cluster to achieve three-petaflop performance. A standard Blue
Gene/P supercomputer configuration will house 4,096 processors per rack.
For Programmers, Friendlier Interfaces & Application Compatibility Speed
Productivity
There are some key differences between Blue Gene/L and Blue Gene/P
supercomputers. In hardware, the Blue Gene/P supercomputer moves to more
(four vs two) and speedier (850 MHz vs 700 MHz) processors per chip;
more memory and an SMP mode to support multi-threaded applications. This
new SMP mode moves the Blue Gene/P system to a programming environment
similar to that found in commercial clusters. The Blue Gene/P
supercomputer dramatically scales up collective network performance to
minimize common bottlenecks inherent in large parallel-computing
systems. Software marks the third key upgrade for the Blue Gene/P
solution -- system management, programming environment and applications
support have all been refined in Blue Gene/P.
(snip)
The Blue Gene supercomputer operating system is based on the open-source
Linux operating system. Applications are written in common languages
such as Fortran, C and C++ using standards-based MPI communications
protocols. The Blue Gene/P supercomputer is compatible with the diverse
applications currently running on the Blue Gene/L supercomputer,
including leading research in physics, chemistry, biology, aerospace,
astrophysics, genetics, materials science, cosmology and seismology.
A variety of independent software vendors have plans to port existing
tools and applications to the Blue Gene/P supercomputer. These include
Gene Network Sciences, TotalView Technologies, Inc., Tsunami Development
LLC and Visual Numerics, developers of IMSL.
For more information, go to:
http://www-03.ibm.com/servers/deepcomputing/bluegene.html
--
Del Cecchi
"This post is my own and doesn’t necessarily represent IBM’s positions,
strategies or opinions.”
Robert Myers
> News release, redacted of puffery
>
> IBM Triples Performance of World's Fastest, Most Energy-Efficient
> Supercomputer
> Tuesday June 26, 8:00 am ET
>
> New Blue Gene/P Designed to Break the "Quadrillion" Barrier
>
http://hardware.slashdot.org/hardware/07/06/26/1517208.shtml
Robert.
Del Cecchi
Robert Myers
> Robert Myers wrote:
> > On Jun 26, 9:10 am, Del Cecchi <cecchinos...@us.ibm.com> wrote:
>
> >>News release, redacted of puffery
>
> >>IBM Triples Performance of World's Fastest, Most Energy-Efficient
> >>Supercomputer
> >>Tuesday June 26, 8:00 am ET
>
> >>New Blue Gene/P Designed to Break the "Quadrillion" Barrier
>
> > Slashdot is all over it
>
> >http://hardware.slashdot.org/hardware/07/06/26/1517208.shtml
>
> > Robert.
>
> Mostly with feeble attempts to get a laugh
>
Hmmm. I'd pretty much given up on actually *reading* slashdot, but,
by sticking to the posts with a score of 5, I got some interesting
information I hadn't known.
Perhaps someone in IBM marketing will even notice the comment from a
"parallel programmer" asking if the processors could be had in small
quantities. I think I asked a similar question almost from the very
beginning. A machine built from x86 may not be the most power-
efficient, but the pool of programming talent is deep. Blue Gene
essentially limits the players to national labs and the like (to
someone who needs at least a rack of hardware that is of limited
usefulness for general purpose computing).
Robert.
Del Cecchi
"Robert Myers" <rbmye...@gmail.com> wrote in message
news:1182993938.1...@k29g2000hsd.googlegroups.com...
presume that small, single board, systems are coming. That is just my
theory of course. But I suppose if there were a lot of demand someone
might listen.
del
Wes Felter
> Perhaps someone in IBM marketing will even notice the comment from a
> "parallel programmer" asking if the processors could be had in small
> quantities. I think I asked a similar question almost from the very
> beginning. A machine built from x86 may not be the most power-
> efficient, but the pool of programming talent is deep. Blue Gene
> essentially limits the players to national labs and the like (to
> someone who needs at least a rack of hardware that is of limited
> usefulness for general purpose computing).
Given that BG/P programming is just MPI with 4-core/4GB nodes, you could
easily learn it on an x86 cluster. It's not clear that having a deskside
version would be that useful. But there is a tiny BG/L available in
Japan, so maybe there will be an equally small BG/P some day.
Wes Felter - wes...@felter.org
Thomas Womack
Del Cecchi <delcecchi...@gmail.com> wrote:
>They eventually did something like that for blue gene/L so I would
>presume that small, single board, systems are coming.
Have you got a reference for the availability to mortals of the
single-board Blue Gene/L?
Tom
Piotr Wyderski
> Four IBM® (850 MHz) PowerPC® 450 processors are integrated on a single
> Blue Gene/P chip. Each chip is capable of 13.6 billion operations per
> second. A two-foot-by-two-foot board containing 32 of these chips churns
> out 435 billion operations every second, making it more powerful than a
> typical, 40-node cluster based on two-core commodity processors.
> Thirty-two of the compact boards comprise the 6-foot-high racks. Each rack
> runs at 13.9 trillion operations per second, 1,300 times faster than
> today's fastest home PC.
How did they calculate that? 32 (chips per board) * 32 (boards per rack)
* 4 (cores per chip) = 4096 cores @ 850 MHz, which is 3.4816*10^12
aggregated cycles per second. The fastest Core2 is now 2.93 GHz, so it
gives 5.86*10^9 cycles per second, thus it is "only" 594 times faster. So
then, does IBM claim that their PPC450 has 2.18 times higher instruction
issue rate compared to a modern Core2 chip?
I can't find anything official about the chip -- does it have AltiVec?
And the standard test: is a single BlueGene rack cheaper than 593 Core2-s?
:-)
> The Blue Gene supercomputer operating system is based on the open-source
> Linux operating system.
Not on the IBM's ultra-professional AIX operating system? Wow.
Best regards
Piotr Wyderski
Chris Thomasson
message news:f663v4$n7e$1...@news.dialog.net.pl...
> Del Cecchi wrote:
[...]
> And the standard test: is a single BlueGene rack cheaper than 593 Core2-s?
> :-)
IMVHO, if we can wait patiently for 7-10 years, we will possible get to
experience 1500+ cores on a single chip...
http://groups.google.com/group/comp.arch/browse_thread/thread/f66494ee093ec9dd
Am I off my fuc%ing rocker?
:^0
Chris Thomasson
news:QJ6dnSkoFo-ndRvb...@comcast.com...
> "Piotr Wyderski" <wyde...@mothers.against.spam-ii.uni.wroc.pl> wrote in
> message news:f663v4$n7e$1...@news.dialog.net.pl...
>> Del Cecchi wrote:
> [...]
>> And the standard test: is a single BlueGene rack cheaper than 593
>> Core2-s? :-)
> [...]
>
> IMVHO, if we can wait patiently for 7-10 years, we will possible get to
> experience 1500+ cores on a single chip...
If the chip designed are as smart as I think they are, well, perhaps we
could get up into the tens-of-thousands of cores per-chip...
Nah...:
http://groups.google.com/group/comp.arch/browse_thread/thread/574295430deb430
Is this crazy or what?
Chris Thomasson
> could get up into the tens-of-thousands of cores per-chip...
'designed' is suppose to read as:
'designer's
BTW, I want to thank all of you hardware folk for all of your extremely hard
work, and dedication. Those of you who designed Intel ships allowed a simple
software monkey like me to actually make a fairly decent living in the 80's,
and actually support more than one person who unfortunately hit some
hard-times along the way!
Thanks!
:^)
Piotr Wyderski
> IMVHO, if we can wait patiently for 7-10 years, we will possible get to
> experience 1500+ cores on a single chip...
I don't think so. Even now it is extremely hard to shrink
the component sizes and I doubt it will be easier in the
future. Moreover, there are serious problems related
to heat dissipation and powering the chip. 1500 cores
means that a single one must use at most 0.1 Watt,
which is science fiction. :-) You may easily achieve that
goal if you redefine the meaning of "core", but if it is
a full-blown 32+ bit processor with a bunch of SIMD
extensions running at ~2GHz, then it will be a pretty
hopeless task to build such a beast.
BTW, what would you do with this kind of device?
Even lock-free doesn't scale up so high, so the only
solution is to use stream processing, which cannot
be applied to many practical tasks. IMHO there
is no reason to cross the 64 core boundary in a
UMA-like configuration. It might be perfectly doable,
as the multi-GHz Pentium4 processor was, but it
seems that Intel has learnt the lesson...
Want n*10^4 "cores", buy a large FPGA chip. :-)
> Am I off my fuc%ing rocker?
Well... ;o)
Best regards
Piotr Wyderski
Chris Thomasson
> Chris Thomasson wrote:
>
>> IMVHO, if we can wait patiently for 7-10 years, we will possible get to
>> experience 1500+ cores on a single chip...
>
> I don't think so. Even now it is extremely hard to shrink
> the component sizes and I doubt it will be easier in the
> future. Moreover, there are serious problems related
> to heat dissipation and powering the chip. 1500 cores
> means that a single one must use at most 0.1 Watt,
> which is science fiction. :-) You may easily achieve that
> goal if you redefine the meaning of "core", but if it is
> a full-blown 32+ bit processor with a bunch of SIMD
> extensions running at ~2GHz, then it will be a pretty
> hopeless task to build such a beast.
I guess you could always resort to increasing the physical size of the chip
itself. A super-computer on a chip, well, the chip size could be 4x4 inches
or bigger sitting on top of memory /w some sort of integrated liquid cooling
system and fans... If somebody complains about the size and extreme cost
factors, well, you can say "of course its big, its a super computer! ;^)"
lol. Fantasy land.
>> Am I off my fuc%ing rocker?
>
> Well... ;o)
:^)
Chris Thomasson
> "Piotr Wyderski" <wyde...@mothers.against.spam-ii.uni.wroc.pl> wrote in
> message news:f683kf$s5c$1...@news.dialog.net.pl...
>> Chris Thomasson wrote:
>>
>>> IMVHO, if we can wait patiently for 7-10 years, we will possible get to
>>> experience 1500+ cores on a single chip...
>>
>> I don't think so. Even now it is extremely hard to shrink
>> the component sizes and I doubt it will be easier in the
>> future. Moreover, there are serious problems related
>> to heat dissipation and powering the chip. 1500 cores
>> means that a single one must use at most 0.1 Watt,
>> which is science fiction. :-) You may easily achieve that
>> goal if you redefine the meaning of "core", but if it is
>> a full-blown 32+ bit processor with a bunch of SIMD
>> extensions running at ~2GHz, then it will be a pretty
>> hopeless task to build such a beast.
>
> I guess you could always resort to increasing the physical size of the
> chip itself. A super-computer on a chip, well, the chip size could be 4x4
> inches or bigger sitting on top of memory /w some sort of integrated
> liquid cooling system and fans...
The memory would be integrated into the ridiculously large chip.
Del Cecchi
I recall seeing the announcement. Sorry I can't be more specific due to
limited connectivity
Chris Thomasson
> Chris Thomasson wrote:
>
>> IMVHO, if we can wait patiently for 7-10 years, we will possible get to
>> experience 1500+ cores on a single chip...
> BTW, what would you do with this kind of device?
[...]
Dynamic navigation system for space travel of 50+ years from now...? Imagine
science fiction spaceship jump into deep space beyond all of our known star
charts and we need computer that can possible predicts where might be in
space and time relative to our initial jump origin.
Chris Thomasson
> "Piotr Wyderski" <wyde...@mothers.against.spam-ii.uni.wroc.pl> wrote in
> message news:f683kf$s5c$1...@news.dialog.net.pl...
>> Chris Thomasson wrote:
>>
>>> IMVHO, if we can wait patiently for 7-10 years, we will possible get to
>>> experience 1500+ cores on a single chip...
> [...]
>
>> BTW, what would you do with this kind of device?
>
> [...]
BTW, there are lock-free applications that can scale up to such devices. You
have to look into lock-free highly-distributed algorithms.
Piotr Wyderski
> Dynamic navigation system for space travel of 50+ years from now...?
> Imagine science fiction spaceship jump into deep space beyond all of our
> known star charts and we need computer that can possible predicts where
> might be in space and time relative to our initial jump origin.
You must be kidding. Have a look at the NASA navigation systems,
they use a pretty ancient technology, but radiation hardened etc. A
computer comparable to Commodore 64 was more than enough to
visit the Moon and return home. :-)
Best regards
Piotr Wyderski
Piotr Wyderski
> BTW, there are lock-free applications that can scale up to such devices.
> You have to look into lock-free highly-distributed algorithms.
Of course such algorithms do exist, but they are more
distributed than lock-free, so I don't count them, as it
is a completely different paradigm. I use the term
"lock-free" only when speaking about shared-memory
multiprocessors...
Best regards
Piotr Wyderski
dave
--
Don't Worry. Be Happy.
Nick Maclaren
In article <1-qdnUeROp4k3hPb...@comcast.com>,
dave <d...@amd.localhost.comcast.net> writes:
|>
|> There is a lot of evidence that the Apollo Moon Flights were faked.
Nah, the whole world is faked. We're fooling you, and there isn't
anything here.
Regards,
Nick Maclaren.
Joe Pfeiffer
No, they didn't have anything near the computing power of a C64.
Joe Pfeiffer
> Piotr Wyderski <wyde...@ii.uni.wroc.pl> wrote:
> >
> > You must be kidding. Have a look at the NASA navigation systems,
> > they use a pretty ancient technology, but radiation hardened etc. A
> > computer comparable to Commodore 64 was more than enough to
> > visit the Moon and return home. :-)
> >
> There is a lot of evidence that the Apollo Moon Flights were faked.
Only when you're looking through the eyeholes of your tinfoil hat.
Chris Thomasson
news:1-qdnUeROp4k3hPb...@comcast.com...
Such as. Humm, are you referring to an instance where you have to squint in
order to make out the stars in the "deep-space" background of some of the
video taken by the USA from the face of the moon?
Chris Thomasson
> Chris Thomasson wrote:
>
>> Dynamic navigation system for space travel of 50+ years from now...?
>> Imagine science fiction spaceship jump into deep space beyond all of our
>> known star charts and we need computer that can possible predicts where
>> might be in space and time relative to our initial jump origin.
>
> You must be kidding.
:^)
> Have a look at the NASA navigation systems,
> they use a pretty ancient technology, but radiation hardened etc. A
> computer comparable to Commodore 64 was more than enough to
> visit the Moon and return home. :-)
Well, the moon is within our grasp.
I was thinking of jumping beyond our known star charts.
Chris Thomasson
news:f6lcef$3q3$1...@atlantis.news.tpi.pl...
The term lock-free can apply to basically any algorithm that does not make
strict use of mutual-exclusion techniques; IMVHO at least... For instance
here is a highly-distributed memory allocator I invented which can scale up
to basically any amount of threads you can imagine:
http://groups.google.com/group/comp.arch/browse_frm/thread/24c40d42a04ee855
Any thoughts?
Chris Thomasson
<basic science fiction book ;^)>
Imagine craft jumping to a another solar-system governed by the gravity-wave
that is emitting from the center of an unexplored galaxy. All of the light
we are receiving is old, "so-to-speak"... The photons have travel long
distances, and are fairly "easily" influenced by large pools of gravity...
What we "currently see" could be vastly old information indeed. All of a
sudden, a genius from some university discovers how to manipulate
space-and-time in a fashion that is perfect for space-travel. How do be
navigate large jumps into space-and-time? Tens-of-thousands of light-years
in an instance, per-jump...
</basic science fiction book>
Eric P.
>
> <basic science fiction book ;^)>
> Imagine craft jumping to a another solar-system governed by the gravity-wave
> that is emitting from the center of an unexplored galaxy. All of the light
> we are receiving is old, "so-to-speak"... The photons have travel long
> distances, and are fairly "easily" influenced by large pools of gravity...
> What we "currently see" could be vastly old information indeed. All of a
> sudden, a genius from some university discovers how to manipulate
> space-and-time in a fashion that is perfect for space-travel. How do be
> navigate large jumps into space-and-time? Tens-of-thousands of light-years
> in an instance, per-jump...
> </basic science fiction book>
Alcubierre warp drive
http://en.wikipedia.org/wiki/Alcubierre_Drive
IIRC unfortunately it requires large amounts of negative energy
to build one. Since no one is quite sure just what negative energy
is or where to find it, it could be a while before you see starships.
Eric
Chris Thomasson
news:468E9032...@sympaticoREMOVE.ca...
> Chris Thomasson wrote:
>>
>> <basic science fiction book ;^)>
>> </basic science fiction book>
>
> Alcubierre warp drive
> http://en.wikipedia.org/wiki/Alcubierre_Drive
>
> IIRC unfortunately it requires large amounts of negative energy
> to build one. Since no one is quite sure just what negative energy
> is or where to find it, it could be a while before you see starships.
please note the clause:
"science fiction" in the wrapper around my post!
;^)
Chris Thomasson
I was wondering if a dynamic navigation system could be designed to
facilitate that type of crazy space-travel...
;^)
Eugene Miya
Chris Thomasson <cri...@comcast.net> wrote:
>>> IMVHO, if we can wait patiently for 7-10 years, we will possible get to
>>> experience 1500+ cores on a single chip...
>message news:f683kf$s5c$1...@news.dialog.net.pl...
>
Is it radiation hardened and space qualified?
>science fiction spaceship jump into deep space beyond all of our known star
>charts and we need computer that can possible predicts where might be in
>space and time relative to our initial jump origin.
It's still fiction.
--
Eugene Miya
Naw, Nick is the woman in the red dress....
And there's isn't a socket in the back of your head.
--
Eric Smith
> Naw, Nick is the woman in the red dress....
One of these lives has a future...