Blue Gene/L Tops Its Own Supercomputer Record

DIY News writes "Lawrence Livermore National Laboratory and IBM unveiled the Blue Gene/L supercomputer Thursday and announced it's broken its own record again for the world's fastest supercomputer. The 65,536-processor machine can sustain 280.6 teraflops. That's the top end of the range IBM forecast and more than twice the previous Blue Gene/L record of 136.8 teraflops, set when only half the machine was installed."

Comments

Beowolf cluster

[totallygeek]

Imagine a Beowulf cluster of...oh, nevermind.

Re:Beowolf cluster

[HawkingMattress]

Sooo redundant, i mean even in this context it has been done a thousand times.

Re:Beowolf cluster

[Pepsi__Blue]

Actually, PS3s will have 1.8TFlops, because that is calculated as a theoretical max (i.e. Total processing power of each individual component is added together) and becuase since graphics are an example of parallelizable problems GPUs produce a lot of FLOPS. See FLOPS and Game Consoles for more info.

Re:Beowolf cluster

[Haydn Fenton]

Ah come on. We all knew this would be the fr0st p1st and we all knew it would be modded up, we love our stereotypical, predictable and redundant jokes, it makes slashdot slashdot.
If you can't beat them, join them and get some free 5, Funnys.

In soviet russia, one for I, welcome our beowulf cluster of old korean hot grits in outer space :)

Re:Beowolf cluster

[Haydn Fenton]

Yeah... but can it run Vista?

hmmm

[jigjigga]

lets put folding@home (http://folding.stanford.edu/) on that mother!

Re:hmmm

[DrEldarion]

Seriously, it's going to be used for nuclear weapons simulations and whatnot. I'd much rather see it going towards protein folding than that.

Re:hmmm

[thesupraman]

While I know that you are joking, one of the major targets of this particular machine is actually basically that, not of course for any direct public benefit, but for the owners.

This particular machine is of course targeted at LANL, and weapons development (oops, did I say that? I mean 'stockpile stewardship')

However, protein folding is one of the primary targets of the architecture.

Oh, and BTW, the IO nodes of this beast run linux. Not exactly a standard kernel, but not far off. The compute nodes run a very simple custom kernel to minimise resource use (after all, they have very limited needs as the IO nodes provide them most services).

Re:hmmm

[CuteVlogger]

Because it's at IBM, it's possible that it'll be used for their World Community Grid at some point, probably when it's idle... The current project there is, in fact, protein folding.

Re:hmmm

[QuantumG]

I think this is more appropriate.

Re:hmmm

[Tuross]

I was thinking more along the lines of... lets put Slashdot on that mother so we might be closer to getting stories that a "news" and not "olds" ;)

Re:hmmm

[Burz]

Let's see, the entire multi-year experiment at ClimatePrediction.net could be completed in about... oh 13 days. :^)

(Not really; I made that up. But if you're curious about how much crunching power we have on tap, visit the project website ;) ).

Re:hmmm

[deglr6328]

"This particular machine is of course targeted at LANL, and weapons development (oops, did I say that? I mean 'stockpile stewardship')"

Just to expand on that, it is worth noting that the ASCI Blue Pacific supercomputer at LLNL was the first to run a fully three dimensional simulation of a nuclear trigger (plutonium fission) implosion and shortly thereafter was the first to run a full 3D simulation of the secondary fusion stage in a thermonuclear device. This computer was capable of ~3 teraflops and took something like 20 days to run those sims. Blue Gene is ~100 times faster than that computer and judging from the time it took ASCI White (~10 Tflops) to complete a simulation of a full thermonuclear detonation, it would therefore probably not be unreasonable to assume this new computer is capable of full 3D simulation of a complete thermonuclear bomb detonation (primary and secondary) in mere hours to a couple days. It is a shame that we even "need" nuclear weapons, but if we're going to have them I for one would much rather see tests of them done in silicon instead of in a big mushroom cloud!

Yes, it is also sad that while other countries use thier supercomputing power mostly to investigate protien folding and earthquake propagation and other purposes generally recognized as peaceful we mainly use ours for simulation nuclear weapons designs; but it is not all bad. The simulations of imploding fusion fuel can (and will) also be used to simulate the implosion of the tiny fusion microcapsules which are imploded in laboratory laserfusion facilities like NIF. This has the potential to eventually result in laserfusion (inertial confinement fusion) as a power source. Supercomputers which were mainly intended to be used for weapons research in the past have occasionally also served up a few surprises in completely unrelated fields. The supercomputer Cray X-MP (?) at Sandia (?) labs in the mid 80s was where the first simulations of the giant impact theory of the formation of the moon were validated. Its now the predominant theory of the moon's origin. It is hard to imagine that this new computer won't have a few surprises of its own to reveal even if it only donates a small amount of time to non-defense related research.

Re:hmmm

[VENONA]

Yeah, right. That's very sensitive of you, I'm sure.

We don't do chemical or biological warfare. All we have is nuclear weapons. So how do we respond, if and when some nutcase state hits us with a nuclear, biological, or chemical attack? Harsh language? Throw some nicely-folded proteins?

Personally, I prefer that we have some assurance that our nukes will work if we ever need them.

You can make any sort of argument you'd care to about our messed up foreign policy. I'd probably agree with quite a bit of it. But I still want our stockpiles to work if we ever need them.

It's a harsh world. Sensitivity and political correctness will only take you so far in dealing with it.

Re:hmmm

[LordFnord]

> Oh, and BTW, the IO nodes of this beast run linux

Yeah? Hmmmm.

lordfnord@eris:~$ ssh bluegene-l.ibm.com

Welcome to Linux 2.6.14

bluegene-l login: falken
Password: joshua

Greetings, Professor Falken. Would you like to play a game?

1. Checkers
2. Chess
3. Protein folding
4. Global thermonuclear war

Uh-oh.

Re:hmmm

[troc]

Greetings, Professor Falken. Would you like to play a game?

Am I the only one who is incapable of reading that (even in the dubious privacy of my own head), without saying it in a "computer" voice?

Troc.

Re:hmmm

[Silverstrike]

Because, when you're in agony dying of radiation poisoning, its really going to make your day a lot better to be assured that somewhere else in the world another person, equally as removed from the political context of the nuclear conflict as you are, will be in just as much agony as you are.

Now, just maybe, the presence of these weapons can be called a deterant, so its possible that possessing them is a necessary evil. However, to be quite honest, if we ever "need" them -- I really do hope they fail to work.

Re:hmmm

[Anonymous Coward]

Actually, I am really a vengeful, sadistic person. Knowing my own nukes are going to kill my enemies (even in a loose sense) would make me feel better as I died from radiation poisoning.

Re:hmmm

[Somegeek]

I think you're failing to take into account the fact that the top Blue Gene/L just doubled its performance, (the subject of the story)?

My totals (with the new Blue Gene figures) show 355.5 US classified vs. 212.8 US non-classified, a fair advantage for weapons research.

Also, this is assuming that we know about all of the classified supercomputers, which seems to be a bit of a stretch for me, but maybe I've read too many spy novels.

Re:hmmm

[andreyw]

Maybe if we didn't spend funds and lives carving out an empire for a certain little country in the middle east, we wouldn't have to constantly live under the threat of pissing the rest of the world and having to deal with nukes, eh?

No that makes too much sense.

Re:hmmm

[Hiro Antagonist]

I agree. I just want the nukes designed such that they only work when someone with a brain 'pushes the button'.

Re:hmmm

[Anonymous Coward]

Yes, it is also sad that while other countries use thier supercomputing power mostly to investigate protien folding and earthquake propagation and other purposes generally recognized as peaceful we mainly use ours for simulation nuclear weapons designs; but it is not all bad.

One might consider that other countries can use their supercomputing power for such purposes precisely because the US uses devotes a relatively larger percentage of its computing power to weapons development. My god is it any secret at all that the rest of the free world has been "free-riding" on US military power since the end of WWII?

Now I'm not complaining about this, or suggesting that we should have thrown them to the wolves, or even suggesting that they're anything less than grateful. But I do get tired of hearing about how militaristic the US is "compared to other [more enlightened] nations" when the US largely provided the very freedom they have to devote significant budgetary resources to matters other than self-defense.

Re:hmmm

[trentblase]

lets put folding@home on that mother!

Since "folding@home" uses distributed processing to put supercomputer tasks on the home computer, wouldn't running the program on a supercomputer just make it "folding"?

Re:hmmm

[lemaymd]

They're upset with us because we're free and have a different ideology. According to UBL (IIRC from the national geographic special), the difference between us is that America loves life and they love death. I hope we never back down from that to appease the other side of the world.

Re:hmmm

[RicktheBrick]

You are assuming that we will know who is responsible for the wmd attack. Do you really trust Pakistan or North Korea or Iran to control all of its weapons. The day may come when a bomb is smuggled into this country or maybe set off on our coastline and we will not have a clue as to who is responsible. The only way we can be safe is to rid the world of all these weapons and the ability to make them. The only way to do this is to first get rid of our own as a moral example to the rest of the world. We could still destroy the economy of any nation with conventional weapon alone.

Re:hmmm

[vertinox]

It's a harsh world. Sensitivity and political correctness will only take you so far in dealing with it.

I dunno... I think the President of Iran could have used a bit more tact lately. Now since he wasn't politically correct, Israel will most likely be dropping smart bombs on their fledgling nuclear facilities.

Case in point, you may look better to your own people/supporters when you aren't politically correct...

But when you piss other people off that are more powerful than you then they tend to "throw chairs" in your general direction. If you are the most powerful nation, you still have to keep in mind if you piss off the entire world they'll start to gang up on you.

I'm not saying that I support political correctness one way or another, but it is just the nature of global politics.

Re:hmmm

[VENONA]

"You are assuming that we will know who is responsible for the wmd attack."
No, I am not. But I certainly think it's possible.

"Do you really trust Pakistan or North Korea or Iran to control all of its weapons."
No, I don't.

"The day may come when a bomb is smuggled into this country or maybe set off on our coastline and we will not have a clue as to who is responsible."
But the day may also come when we're attacked, and we *do* know who did it. And yes, I do worry that we'll have another Shrub-like creature in office if it should ever happen.

"The only way we can be safe is to rid the world of all these weapons and the ability to make them."
It does not follow that this will make us safe. It may make us less safe. The theory and politics of deterrence have been argued for half a century. I doubt we'll resolve all of those issues on Slashdot just now. My take is that a lot of smart people, with a lot of information available to them, have thought deeply about this problem. They've apparently decided that nuclear deterrence is still viable.

I will never be in a postion to duplicate their effort, and make more than a gut call. So my gut call is to trust the people who've made it a major part of their life's work to understand the issues.

"We could still destroy the economy of any nation with conventional weapon alone."
The country that attacks us may not have much of an economy to destroy. I can picture a nation with serious internal problems, fanatical leaders, and WMDs externalizing their problems, and attacking the US. I'm sure you can too.

Re:hmmm

[brsmith4]

That would be a complete waste of its capabilities. f@h is an embarassingly parallel app that requires very very little in the way of communications. BlueGene was designed to run simulations that require a sound, low-latency, high-bandwidth interconnect for lots of communications. f@h is much better off running on a bunch of PCs like it does now.

I always laugh when i hear people say such things because it alludes to a complete ignorance of the field of scientific computing. Its not a bad thing per se, just rather funny. I don't expect everyone to be a supercomputing expert but the seti@home and folding@home references just have to stop. With both programs, you are dealing with algorithms much better suited to be run on the distributed desktop grid paradigm rather than a tightly-integrated parallel computer that is capable of so much more. There is much more to take into account with regards to parallelism than most think.

What's even funnier is the proclaimed 'insightfulness' of the comment. I would have to disagree and say "humorously uninformed". No offense intended as many people would assume the same thing. Many cluster "n00bs" slap together a beowulf for the sole purpose of running f@h or s@h. It's a nice learning experience, but the programs would be better off running on that pimp Athlon64 box they have under their desks than on a rag-tag group of Pentium 233s sucking down more energy than they do useful work. The same thing really applies. BlueGene can be doing research that actually demands its capabilities rather than running an app that barely scratches the surface of what it has to offer. Just my $0.02.

Re:hmmm

[oliverthered]

Throw some nicely-folded proteins?

Well, giving the people the cure for cancer or genetic diseases may stop them from attacking us in the first place, apply game theory to social systems, not the one with the biggest bomb wins.

Re:hmmm

[rtaylor]

Now, just maybe, the presence of these weapons can be called a deterant, so its possible that possessing them is a necessary evil. However, to be quite honest, if we ever "need" them -- I really do hope they fail to work.
You don't even need to possess the weapons for them to be a deterrant.

Canada has the parts and expertise required to build bombs and a delivery system within about 3 months. Heck, they used to sell the US weapons grade plutonium and many of the designers of various rockets came from the Avro Arrow team after leaving for various US firms.

3 months is plenty of time to notice that something is going very wrong in world politics.

For this reason Canada is considered to be armed country despite not having a single Nuclear WMD.

Canada is in the same boat with biological weapons. All the parts and expertise including the research labs and viral strains necessary but their military has not bothered to actually build one in a while.

Re:hmmm

[pimpimpim]

Most of the protein folding studies really don't gain a lot on these huge systems, you would want to use about 32 processors per job maximum (unefficient load balancing / communication starts hindering scaling above that). It's nice to have the ability to do a lot of 32 proc. jobs at the same time without having to wait, but it will be inefficient to use all the processors at the same time for one job, so this max. teraflop-value isn't of much importance for protein folding studies in practice.

Most of the people doing protein folding studies nowadays could just suffice with access to a moderately big fast-interconnect opteron cluster, really doesn't have to be as huge as those top-500 systems.

I guess that there are a lot of areas (nuclear physics, ab-initio), where you can actually gain speed by adding processors and memory ad infinitum, for these people the huge scalability of these IBM machines will be very useful.

Re:hmmm

[Fordiman]

Actually, ensuring that a nuke will work only takes two machines - one low-end and one high-end.

The low-end one is the ballistic tragectory calculator. That's at home base. Don't get me wrong, it has to know weather patterns, trigonometry, physical law, airline routes, satellite tracking, etc. I think maybe a 500MHz machine could handle it.

The high-end one goes on the missile itself, and handles local tracking, GPS, etc, in real time. Local tracking should be done via passive radar and visual detection (what takes the hard work). GPS is for hitting your target. You've got a bit of leeway, as you're essentially swatting flies with a hand-grenade (cue monty python reference here). Local tracking is for avoiding annoying things like interception missiles, airplanes, etc, that might mitigate the damage done (exploding a nuke in the air may not kill everyone in the expected radius, you know). Probably a 3GHz machine could handle this, but I'd go as low as 2.0.

No, you don't really need a 64k paralell processing behemoth. You just need a paired computer for each missile (they could likely even stay in communication with a matched set of encryption keys. Keep 'em guessing, you know?)

Reader

[Jozer99]

They say it can launch Adobe Acrobat Reader in ELEVEN SECONDS!!!

Re:Reader

[Agret]

Awesome! Wish it ran that fast on mine! Use FoxItReader. It's a lot faster and smalller. Homepage here: http://www.foxitsoftware.com/ Download here: http://www.foxitsoftware.com/pdf/pdfrd.zip

Re:Reader

[danormsby]

Maybe we can use it to write large Word documents?

Re:Reader

[Phat_Tony]

I was going to use:

It even meets the minimum system requirements for Longhorn!

Re:Reader

[TheAdventurer]

Yeah, I was heading in the direction of, "Sweet now I can run Quake 4 and F.E.A.R. at the same time!"

Re:Reader

[robvangelder]

I hate Acrobat Reader's load time too. Here is how to speed that up.

Go to the Acrobat program folder:
eg. C:\Program Files\Adobe\Acrobat 7.0\Reader\

Move all of the files and folders under the "plug_ins" folder to the "Optional" folder
The plug_ins folder should now be empty. Acrobat Reader loads faster.

I don't know what those plugins are for, but my PDFs read fine.

Re:Reader

[diablomonic]

I must thank you too... much appreciated

Re:Reader

[StressedEd]

Hmmm, cunning... Ta.

Re:Reader

[slavemowgli]

Here's another option: go to http://www.cs.wisc.edu/~ghost/ and get GSview (the Ghostscript file viewer), and associate that with PDF files. Admittedly, you'll get a nagging screen when it starts up telling you to register, but it's fast, it works, and maybe most importantly, it does not contain any "phone home" functionality or Javascript-in-documents handling or the like (well, it's unlikely that it does, at least).

I've ditched Acrobat Reader in favour of this, and it's one of the best decisions I ever made.

Re:Reader

[wikkiewikkie]

You can also accomplish the same thing by holding down the shift key while Acrobat is launching. It will prevent the plugins from loading.

Re:Reader

[Zeussy]

Or maybe use a non bloated PDF reader like Foxit Reader

Re:Reader

[darkfrog]

And it still can't run Doom3 at full specs!

Re:Reader

[Skuld-Chan]

Then someone will call adobe technical support with complaints like "I can't fill out forms in my adobe" and "my adobe won't read ebooks anymore"

Re:Reader

[aardwolf64]

Note that by moving ALL of the files and folders, you'll be disabling the Search functionality in Acrobat Reader.

Re:Reader

[mikapc]

Better yet, stop using adobe reader and download foxit reader. It loads pdfs instantly and only takes less then 10 megs of space.

compiler?

[LetterRip]

I wonder how much could be gained via compiler improvements, anyone know what compiler they use?

Re:compiler?

[Compholio]

I wonder how much could be gained via compiler improvements, anyone know what compiler they use?

Since it runs Linux and PowerPC? Probably GCC.

Re:compiler?

[Aardpig]

On something like this, they would probably be programming in High Performance Fortran or Fortran w/ OpenMP -- or some similar dialect that supports massively parallel execution. I'm sure IBM develop an in-house compiler for the language.

Re:compiler?

[joib]

Actually, the system is provided with the IBM XL family of compilers.

Re:compiler?

[SnowZero]

Don't forget to compile with:
make -j 65536

Re:compiler?

[promatrax161]

Actually I have heard a talk on astrophyisical N-body simualtions which run on 512 POWER4 processors. By using a combination of OpenMP on the nodes and MPI between the nodes you can do reasonably well (depending on how your problem changes with time). IBM's XL compilers are used.

Re:compiler?

[joib]

They can't use OpenMP since they don't have shared memory beoynd 2 cpu:s. HPF is dead, or least dying, due to lackluster scaling beyond a few dozen cpu:s.

What they use in practice is MPI.

Re:compiler?

[Anonymous Coward]

Actually, most machines are partitioned into front end and back end. The back end is for running large production runs (1000's of PEs) and is usually on accessible as a batch queue. The front end is for compiling and debugging and is interactive (perhaps even running serially). The front end might even be another machine.

Contrary to popular /. opinion, compiling is not a big task. Especially when compared to the real calculations done.

Big machines like this usually have another queue on the front end for long compilation jobs.

So make -j 4 might be more appropriate. Unless you wanted to piss other users off.

Re:compiler?

[Anonymous Coward]

In theory, OpenMP allows you to do some nice things you can't accomplish well in MPI - load balancing the work being done on several CPUs on a node being one of the most important. However, in practice, the addition of OpenMP often complicates a code for little benefit because:

  a) If your code load balances on a node but not between nodes, you're still rate-limited in domain decomposition type of problems by the data taking the longest step. There are strategies to minimize this, but when we're talking about codes running on 64K processors, this is already a big worry!

  b) More importantly, the early benefit OpenMP gave over MPI when running on shared memory is no longer there! Most modern MPI distributions can optimise communication from one CPU to another (via MPI calls) that happen to be on the same node to use shared memory segments instead of sending the message out to the wire.

    The added complexity of hybrid models are rarely worth it in these days of (generally speaking) large numbers of small-way SMP nodes. Perhaps systems like NASA's 20 x 512-CPU system are more amenable to such things, and with the upcoming Horus chipset for up to 32-way Opterons, this could see a revival, but MPI is by far the dominant paradigm, and rightfully so.

Re:compiler?

[gnuLNX]

Actaully would need to install distcc and probably do somthing like:

make -j 65536 CXX=distcc

Re:compiler?

[slashflood]

You should always set the number of parallel jobs to CPUs+1. In this case make -j 65537.

Re:compiler?

[EMH_Mark3]

Somehow I don't think that rule is entirely correct for 65000+ cpus.

Re:compiler?

[Nethemas the Great]

*boggles at the concept*

emerge x11-base/xorg-x11

Re:compiler?

[slashflood]

Somehow I don't think that rule is entirely correct for 65000+ cpus.

Somehow I think that my grandparent post should be modded funny.

Perhaps we could use it to..

[278MorkandMindy]

..figure out what the hell we are going to be doing for energy in 15 years??

"Look to the future and the present will be safe"

Re:Perhaps we could use it to..

[John Hurliman]

And moments before the breakthrough is finally computed, a blackout occurs due to the power consumption of the supercomputer.

Re:Perhaps we could use it to..

[markus_baertschi]

Actually one of the projects running on the BlueGene at EPFL is to simulate plasma turbulences in magnetical fields. This is part if the international ITER project for the fusion reactor to be built in south France.

Markus

For various uses

[strider44]

An IBM engineer was caught remarking "And boy can it hold a lot of porn."

Re:For various uses

[porkThreeWays]

>>Blue Gene/L and ASC Purple together are expected to consume 10 megawatts of the 45-megawatt capacity that LLNL's Terascale Simulation Facility can supply for computing and cooling.
Sadly, the mcfly computer won't get off the ground until it can get it's 1.21 gigawatts.

That's a bloody fast supercomputer...

[kyle90]

The damn thing's smarter than I am. Well, that's taking an estimate of 100 teraflops for the human brain, which seems to be popular.

Re:That's a bloody fast supercomputer...

[andymar]

It's all hardware. To be smarter than you it also needs software, preferably emulation of the human brain. It will be many years before we can do that.

Re:That's a bloody fast supercomputer...

[Quirk]

Here's a quick rundown on the numbers. Brain Computing

Re:That's a bloody fast supercomputer...

[QuantumG]

An annoying rundown. I hate it when people don't just estimate the number of ops outright. 3.0 x 10^17 integer operations per second is the estimate. Why integer, I have no idea, as every neural network I've ever simulated used floating point operations, but that's probably because I wasn't aware of some brilliant optimisation (doh!). But whatever, let's pretend that number is accurate for floating point operations. 280.6 teraflops = 280.6 x 10^12 or 2.806 x 10^14, so you need over 1000 of these new computers (or you need to run at 1000th of realtime, hmmm). Then you need to actually determine an algorithm to simulate such a huge neural network on a distributed platform. Then you need some magical way to scan a human brain. Actually, that might not be too hard. If you zap someone who is clinically braindead they come back to life and seem normal enough (although they seem to consistently be more religious.. hmm, that might not be such a good thing) so maybe all you need to do is grab the brain of someone recently diseased and slice layer by layer from front to back and scan it with an electron microscope. That should give you a pretty good map. From simulating small parts of the map at a time you should be able to learn a lot. At least enough to provide it with input and output for a virtual environment.

Re:That's a bloody fast supercomputer...

[Quirk]

...so maybe all you need to do is grab the brain of someone recently diseased and slice layer by layer from front to back and scan it with an electron microscope. That should give you a pretty good map. From simulating small parts of the map at a time you should be able to learn a lot. At least enough to provide it with input and output for a virtual environment.

The above recalled Rudy Rucker's early, first(?) novel Software. The plot carries your idea along the following lines:

"Cobb Anderson created the "boppers," sentient robots that overthrew their human overlords. But now Cobb is just an aging alcoholic waiting to die, and the big boppers are threatening to absorb all of the little boppers--and eventually every human--into a giant, melded consciousness. Some of the little boppers aren't too keen on the idea, and a full-scale robot revolt is underway on the moon (where the boppers live). Meanwhile, bopper Ralph Numbers wants to give Cobb immortality by letting a big bopper slice up his brain and tape his "software." It seems like a good idea to Cobb."

Rucker's background gave him more insight into the idea of mapping consciousness than "Dixie Flatline's construct" William Gibson came up with.

The ideas propagated in epistemology, (as it refers to the theory of knowledge not the methodology of science), have eaten up big chunks of my time since my mid teens. Presently I'm trying to come a working definition of information, more so in the context of entropy and negentropy, as it applies to life. The going has been slow. Recently I've read through quite a bit of Complexity theory (Santa Fe Institute, S. Kaufman et al) and am poking around at ideas of memory and attractor basins. The same old story the more I learn, the less I know. We, as sentient beings, seem to need to relate through stories and broad context. Factor in Antonio Damasio's work on the importance of emotion in knowing and decision making, and, the idea of modelling consciousness becomes, to me, presently impossible.

cheers, thanks for the input.

Re:That's a bloody fast supercomputer...

[LS]

It's not just hardware or software, it's how the hardware is organized. The brain isn't based around the CPU/memory/bus/instruction set model, so emulating it on a computer will probably result in a loss of efficiency of multiple orders of magnitude, making the terraflop measurements meaningless.

LS

Re:That's a bloody fast supercomputer...

[Mandrel]

Since there's a good chance brain behaviour doesn't depend on computations and timings much shorter than a milisecond, a simulation on a conventional parallel computer need not mimic the brain's layout of computation and communication, but can employ spatial and temporal multiplexing to achieve the same effect. This, coupled with that fact that an architecture with a small fast cache memory and a larger slower main memory is a good match for respectively holding neuron and synaptic states, makes me believe that current supercomputer architectures like Blue Gene would simulate a brain quite efficiently, given the right software.

In fact simulation of the cerebral cortex is one of the tasks Blue Gene will undertake

Re:That's a bloody fast supercomputer...

[Anonymous Coward]

Some of the replies below comment on the operations-per-second estimates of the brain, so I won't. There is, however, a really interesting project at the Brain Mind Institute in Switzerland to simulate, in large detail, a full neocortical column of the brain on several BG/L racks, donated by IBM for this cause. As a consequence, this project is called the "Blue Brain Project", and you can read a bit more about it here:

IBM
http://domino.research.ibm.com/comm/pr.nsf/pages/n ews.20050606_CognitiveIntelligence.html

New Scientist
http://www.newscientist.com/article.ns?id=dn7470

Main site (which doesn't seem to be working at this moment!)
http://bluebrainproject.epfl.ch/

Re:That's a bloody fast supercomputer...

[diegocgteleline.es]

Well, may be you, but it's certainly not smarter than me - I can write code, for instance.

Re:That's a bloody fast supercomputer...

[SatanicPuppy]

The interesting thing about human consciousness is how it seems to arise specifically from the interaction of supremely primitive internal stimuli (hunger,etc) and incoming sense data.

I'm not sure the mechanics of intelligence matter so much as a stimulus response pattern fed by external data collected in repsonse to an internal drive. We like to think there is magic stuff going on with our minds, but we're really just pattern users, and those patterns are ingrained in the most primitive of ways.

Seems like hardware having only one non-distinguishing "sense" would be a problem...Humans learn very early that certain things are "good" (eg pleasurable/non-painful) and to be sought, and some things are "bad" (eg painful) and to be avoided. But all that depends on stimuli in the world, and since the only stimuli available to a computer would seem to be data, and since it seems to be difficult to define pleasureable/painful data...

Re:That's a bloody fast supercomputer...

[LS]

I'm not an expert on the brain, but isn't it common knowledge that we do not yet fully understand it's workings, let alone the physics and chemistry that underly the atoms the make it up? We can only simulate to a certain aproximation, and I'm not confident that all the conscious and behavioral phenomena the brain exhibits can be explained or simulated without a full understanding of how it works. Considering that brain essentially creates subjective reality (everything you are thinking, seeing in front of you, smelling, etc.), I think that it's complexity is deep, perhaps infinite. We do not yet know how deep the complexity of the brain, or the universe for that matter goes.

LS

Re:That's a bloody fast supercomputer...

[theurge14]

Then why can't I remember what I ate for lunch three days ago? Do I need a RAM upgrade?

Also, why do I always struggle factoring equations? Do I need an OS upgrade?

Seems like an awful waste of processing power... :(

Re:That's a bloody fast supercomputer...

[Frazbin]

Unfortunately, that estimate neglects any internal processing capabilities that neurons might (do) have. It's generally thought these days that neurons do more than simply sum their input signals-- they've been observed doing much stranger things. Keep in mind that every neuron is a complete (though specialized) cell, and cells are awfully complicated little buggers. To *accurately* simulate the brain, you'd have to multiply the estimate by the average number of operations it *actually* takes to simulate whatever processing a neuron does. It may turn out that the things electrical signals in neurons provoke chemical reactions (which in cells, inevitably involve proteins) that change the output.

In other words, if I gave you 3.0 x 10^17 operations per second and a model of the synaptic connections between all the neurons of a human brain... Well, you would get some damned interesting results, but it wouldn't be anything remotely human. Before you can simulate a human brain in an accurate way, you have to account for internal processing and the (none too simple) action of neurotransmitters.

Nevermind if you want to simulate, say, the brain of someone consuming a chemical with unknown effects but a known chemical makeup (one very useful application for a complete electronic model of a human brain). Then you have model the structure of every type of receptor and determine their concentrations on a neuron by neuron basis throughout the brain-- then talk about the affinity of the chemical for those receptors.

As you can see, talking about two operations per second per synapse is extremely unrealistic, so the numbers we get for a computer powerful enough to simulate a human brain are absurdly low.

Re:That's a bloody fast supercomputer...

[FirienFirien]

Smart? Depends on your definition of "smart".

You can put a whole bunch of things in parallel - I'm sure computers with the 'intelligence' of a texas instruments scientific calculator could be paralleled hugely to create something that does even more teraflops... it just wouldn't be in a useful way. I know that analogy is flawed, but my point is that even Blue Gene/L is useless and dumb until someone smart comes along and puts a distilled level of their smartness (ie writes a program that mimics a process the person can do, just faster) onto it.

The damn thing is faster than the maximum speed of the human brain. Now THAT's saying something. The brain can control the individual muscles (in patterns it has practised) in precise and determined ways to run across a field and catch or throw a ball, while listening to and comprehending motions of atoms in the air (noise, speech (complicated!), changes in amplitude and understanding where multiple things are and how fast they're going in which directions), interpreting however many million pixels we get from the eyeball (and realising which ones are important and which ones aren't) while also negotiating the terrain and knowing where to get to to catch the frisbee and knowing how to catch it and and and etc etc - and can rapidly teach itself to learn something new that it can transfer and apply correctly to a previously unknown situation. That's "smart". The computer is able to do something it's been told how to do, over and over again, very very fast. It can learn, but only if a lot of effort has gone into teaching it how to learn, and it can only learn within the constraints visualised by the programmer.

By having huge numbers of layers of understanding between the interface and the machine code, the supercomputer can do 'smart' things incredibly fast. But it itself is not smart.

Yet >:)

Re:That's a bloody fast supercomputer...

[FirienFirien]

I guess you can argue that at a given time a computer of any size will have a certain pattern of 0s and 1s that will cause it to behave in a predictable way (even though the prediction itself might take an incredibly long time to figure out), but we still have no clue whether that's true of biological systems too so I thought I'd leave that one out (and that's before we even get to things like http://www.netscrap.com/netscrap_detail.cfm?scrap_ id=73 where disconnected circuits can still affect the outcome by induction with nearby circuits...)

Re:That's a bloody fast supercomputer...

[Mandrel]

Just as the once inexplicable, deity-assigned mysteries of the movement of heavenly bodies and the origin of complex life have been substantially explained, I don't think we should fall into believing that a explanation of the workings of the brain will forever be beyond us. There has been substantial and steady progress over the past twenty years — we know more about how the brain is structured and operates than you may think. I think we're close to pulling it all together. There's no reason to believe exotic physics will be part of the solution.

You may think it's a conceit to believe all this, but I think it's just as much a conceit to believe that human thinking is inexplicable.

Re:That's a bloody fast supercomputer...

[drinkypoo]

Integer operations? The brain is analog, you can't even accurately model its behavior with integers, OR floating point numbers, you can only get progressively closer as you get more precise. And, humans are not very good at doing math anyway. I mean, not compared to, say, a game boy.

Re:That's a bloody fast supercomputer...

[LS]

Ok, I'll be honest and admit that I do think it's a conceit to believe that we are close to pulling it all together. But I don't also think it's inexplicable. All I'm saying is that if we can't yet see the light at the end of the tunnel, how do we know how far the tunnel goes?

gentoo...

[dethmetaljeff]

Is it bad if the first question that came to mind was "I wonder how quickly I could install Gentoo on that?"

Re:gentoo...

[B3ryllium]

It would take about four days.

But on the plus side, it would only be, like, three commands.

Re:Results?

[Anonymous Coward]

What useful science has the "Earth Simulator" produced?

What about 42?

I have just one question for Blue Gene/L

[Anonymous Coward]

and the answer had better not be 42.

Re:I have just one question for Blue Gene/L

[NoGuffCheck]

Your in luck, we just got the print out at it reads 0b101010. ...ahh we have no idea what that means, damn thing must be broken!

Re:I have just one question for Blue Gene/L

[raoul666]

Dear god man, you have the question???!!!???111one

But does it run...

[c_forq]

OS-X? Sorry I'm lame...

Re:But does it run...

[Hiro Antagonist]

Yeah, but you won't be able to use it until Spotlight finishes up...

Re:Results?

[Propagandhi]

What useful science has the "Earth Simulator" produced? What useful science will this monstrosity produce?

I totally agree with the sentiment of your post. Scientists have been doing nothing productive for years now, wasting our tax dollars and patience! It's time to put psychics in charge of all scientific endeavors, that way we can stop wasting time on ones that fail to produce non-earth shattering results!

Whats that? This was made by a private company? Oh.. well.. still, they shouldn't be wasting all this time on frivolous niceties like the world's most powerful super computer. .. ...

Do you ever get tired of chastizing people doing great things from the comfort of your computer chair?

Why'd they get the prediction wrong?

[hansreiser]

The legitimate thing that I can imagine is if it was a cost based contract that was given out before the cost of the hardware was known.

Was it?

Cool

[sheuer]

Back when It was only half installed I got to take a tour of it while it was in Rochester, MN... Got to walk through it and touch it. Turns out the computer that controls blue gene takes up about half as much space as blue gene itself.

Comment removed

[account_deleted]

Comment removed based on user account deletion

Re:speed

[strider44]

What a stupid thing to ask. It's running Linux, you should be asking what the frame rates in Doom 3 are.

Re:Results?

[utnow]

hmm... the world's most powerful computer... extremely expensive... built by a private company...

Yeah... they're definatly losing money on this. businesses love to lose money.

Even if it's just a pissing contest, they're gaining publicity that (they think) is worth the cost.

1983

[Xrathie]

WOULD YOU LIKE TO PLAY A GAME? aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa

Re:still en vogue?

[drkstrm]

If it uses 1.21 gigawatts I'm going to hide and not come out until doomsday! :)

Re:Results?

[tpv]

Whats that? This was made by a private company?

But it was paid for by the US government.

not a compiler issue

[Quadraginta]

I have some very limited experience with this kind of computing, and I don't think the compiler is anywhere near the limiting factor.

I strongly suspect the limiting factor is algorithms. That is, the problem is designing code that can efficiently use a massively parallel machine. It's enormously difficult to even imagine how a problem could be solved by breaking it up into 65,000 mini-problems that can be solved simultaneously, and therefore mostly but not entirely independently. People just don't think that way. (Or rather, they do, but only at such a basic level close to the neurons that they are utterly unaware of how it's done.)

This is one reason "parallel computing" has been the Wave Of The Future(TM) for decades, and exhibits the same kind of "promise" as fusion power -- namely, we are told that ten years from now it will change everything -- and we hear it again every ten years.

Re:not a compiler issue

[wenchmagnet]

Blue Gene/L: Hardware Overview and Planning - http://www.redbooks.ibm.com/abstracts/sg246796.htm l?Open

Blue Gene/L: System Administration - http://www.redbooks.ibm.com/abstracts/sg247178.htm l?Open

Blue Gene/L: Application Development - http://www.redbooks.ibm.com/abstracts/sg247179.htm l?Open

Unfolding the IBM eServer Blue Gene Solution - http://www.redbooks.ibm.com/abstracts/sg246686.htm l?Open

Electricity Bill

[tpv]

I'd be able to figure out how to heck to pay the huge electric bill this would generate.

Easy - you'd run a huge federal deficit, and let future generations sort it out.

Was to be expected

[halleluja]

.. since Quake 4 just hit the shelves.

Super-linear speedup?

[RovingSlug]

more than twice the previous Blue Gene/L record ... set when only half the machine was installed

When it was half done it was less than half the speed? Impressive. Was there a software/OS upgrade along the way, as well?

Re:Super-linear speedup?

[SamMiller0]

I think the MPI stack was improved significantly between the 32 and 64 rack installations.

Re:Results?

[Jeremy Erwin]

What useful science has the "Earth Simulator" produced?

You might try reading The Journal of the Earth Simulator.

Or perhaps this summary of 2003 research

The 2005 projects are listed here

And it's already...

[SnarfQuest]

Only listed a few minutes ago, and it's already been slashdotted...

Picture

[TechnoGuyRob]

Here's a picture of the momma: http://en.wikipedia.org/wiki/Image:BlueGeneL-600x4 50.jpg

Re:Picture

[Anonymous Coward]

whoo! look at all that electrostatic discharges between the data contact points!

martyna

[Quadraginta]

Way off topic, but your link says Glenn Martyna moved to IBM/Watson. Christ, better buy some IBM stock.

By the way, a trivial point with respect to this: Isn't it relativity, not QM, that forbids superluminal communication? I seem to recall non-relativistic QM with instantaneous action at a distance (e.g. Coulomb's Law) being alive and well in the realm of quantum chemistry, or perhaps really anywhere pair creation is not an issue.

Re:martyna

[QuantumG]

Whatever, it's geek humour.

Re:martyna

[Anonymous Coward]

Don't pay much heed to that page, there's so little understanding of QM in there that he'd better been making a joke while writing it. Come to that, QG, you should be marking it as such, least people who don't know better take those 'interpretations' seriously - you can never underestimate the desire of people to believe in things :-)

now, as to your question, the light speed limit on signals is imposed by the assumption that the Lorentz group is a valid symmetry of the universe. Apply that to classical mechanics and you get special relativity; apply it to QM and you get relativistic QM. And relativistic QM is alive and well in atomic physics/chemistry, starting with the electron spin - or, less trivially, with corrections to the electron energy levels that usually involve the fine-structure factor (particularly for inner electrons of heavy atoms, but you can look up the solution of the Dirac equation for the Hydrogen atom). Of course, if those corrections are too small for your purpose (or level of numerical error) then you might as well make your life less complicated and ignore them ;-) Just as you don't compute relativistic corrections for a moving car ... at least not for the near future evolutions of cars.

Re:Results?

[LarsWestergren]

What useful science has the "Earth Simulator" produced?

Yes, better climate models and weather forecasts are obviously not needed. A little rain never hurt anyone, as this years hurricane season clearly shows.

Re:Results?

[catch23]

What useful utility has top sports athletes contributed to society? We probably pay the top athletes the cost of several super computers and all they do with all this money is throw orange plastic balls pumped with air into little string baskets hoisted on a pole. Sometimes I wonder what is the purpose of paying some man to hit a tiny ball with dimples into a hole far far away on a grassy playing field. And every time someone breaks someone else's record of hitting the tiny white ball, they get large sums of cash. But when these super computer people break records, all they get is a little pat on the back. BlueGene/L should be signing contracts with Nike and Pepsi.

GPL

[Yartrebo]

What's so hard about releasing these things under an open source license.

I have nothing against donating CPU cycles, but I have yet to find a group that doesn't require me to sign a restrictive software license. And for this particular project, it's a university running it no less. Aren't universities supposed to encourage the spread of information?

(Then again, I'd have to bury my head in the sand and forget about all the patents that universities have amassed, often using tax dollars to fund the research that led to them).

Re:GPL

[Solra Bizna]

What's so hard about releasing these things under an open source license.

Making a client like this Open Source only makes it that much easier to pump the server full of bogus data. You have people who think vandalism is funny to thank for this.

-:sigma.SB

Re:GPL

[bsartist]

What's so hard about releasing these things under an open source license.

Basic scientific method, really - control the environment as tightly as you can, and then document everything as thoroughly as you can. The first precludes open source while the experiment is ongoing, while the second requires opening up the source once the experiment's done.

Aren't universities supposed to encourage the spread of information?

Accurate information, yes. How would you propose that accuracy could be guaranteed with an open client that anyone could alter?

Oh, and don't bother starting in on how binary-only nodes could be hacked, wires can be tapped, etc. - I know that. It's irrelevant. The goal of an experiment like this is to eliminate any variables other than the ones you're testing. Not every variable can be eliminated, but that's not a legitimate reason to abandon the effort entirely. Besides which, only someone with malicious intent would bother going to that kind of trouble; an open-source client could be comprimised by a well-meaning hacker who tried to "optimize" his copy of the client by taking short-cuts.

Re:GPL

[Yartrebo]

The problem of rogue clients can be caught by doing random spot checks. Have a few jobs done twice. If they disagree, then one (or both) of the two clients messed up. Rerun the job on a university computer and ban the offender. Do checks often on new clients and trust most results coming from established clients.

And yes, I would be one to mess with the code, but I wouldn't use the modified code on the internet, it would be for internal tinkering (since I love to play with simulations and would love to get my hands on a climate simulator). To do that would require source for both client and server, and I'm not about to go asking for permission just to get a toy to play around with (and according the web page, personal entertainment would probably not be a valid reason to get access to the source).

From my experience with open source projects, proper procedure is to submit the optimised code (as a diff file) to the developers so that they may incorporate the code into the codebase as well as do any extra tests they may want to do. Optimizing the code and just running it on your client isn't going to be of much benefit to society.

Lastly, I can see the reason for malicious intent, mostly coming from anti-Kyoto and pro business-as-usual types, who have plenty of resources to hire black hats or maybe even get the CIA on the job.

Re:GPL

[Burz]

I think this suggestion has been made before, and it may a good idea. However the initial response mentioned that the code wasn't owned by the project or the university, but is on-loan from the UK's Met Office. The CPDN-distributed models are even named "Hadley", the Met's climate research center that has developed these models over the course of decades.

You seem prepared to make a well-reasoned argument for open sourcing the models, so you may want to start at the discussion board. Perhaps project members can offer insight on dealing with the Met Office.

Deep Computing Capacity on Demand

[e.loser]

"In comparison, the on-demand program charges about $10,000 per week to use one-eighth of a Blue Gene/L rack" found here

With 1024 processors per rack, and does that mean you only have access to 128 processors? Or only 1/8 of all of the Blue Gene racks? Not very much of a savings, considering 128mil for all the racks, excluding costs for other equipment and not to mention tax.

Re:still en vogue?

[diablomonic]

the article states 10 Mega watts (although I think that was the combined power usage for this and another supercomputer)

Nitpicking

[darkitecture]

WOULD YOU LIKE TO PLAY A GAME? aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa

If you're going to be an 80s geek, don't half-ass it like most people. The correct line from WarGames is "SHALL WE PLAY A GAME?"

I know what I'd do...

[Jack Earl]

If I had a computer that fast...

#include <stdio.h>
int main(){
int answer = 42;
printf("%d", answer);
}

Re:I know what I'd do...

[Jack Earl]

Just pretend return 0; is in there.

Re:I know what I'd do...

[scotch]

I'll just pretend you compile with a c++ compiler instead. Has the implicit 'return 0;' at the end of main been adopted into C99 from C++?

Re:I know what I'd do...

[section321a]

But... what's the question?

Upgrade complete-

[Anonymous Coward]

Just in time for... Windows Vista(tm)

Notice that performance had increased per cpu...

[Anonymous Coward]

Notice that the performance has actually increased PER proccessor as you add more proccessors... This is very remarkable in computer technology.

Normally when you add cpus to a computer you get a increase in performance, but it doesn't increase linearly with each cpu. You have one cpu you have 100% performance, add one more and you may have 180% the performance and add 2 more you may have 300% of the performance etc etc.

Notice that with half the machine there it got 138 GFlops.

So if you doubled the size of the machine you'd expect to get something like 260 Gflops per second.

But you have 280 Gflops per second.

This pretty much means that as you add cpus the performance of each cpu actually increases slightly. That's a exponentional growth rate, at the beginning of the curve.

Of course there has to be a technical limit to the system and the amount of space, heat, and electricity it can handle.. but technically if you double the size of the cluster again I wouldn't be suprised if you'd get close to 750 GFlops per second performance.

This is some seriously hardcore stuff, the future of computing hardware. Todays supercomputer, tomorrow's desktop.. I can't wait.

The run down says:

[Inoshiro]

""Earth Simulator" supercomputer performs 36 Terra flOps / second. ...Earth Simulators required to model 1 brain = 3.0 x 1017 / 3.6 x 1013 = 8333. ...
1 Brain = 8333 State-of-the-art Supercomputers"

So, unlike what kyle90 posted, you'd actually need 1,443 (rounded up from 1,442.25) of these Blue Gene/L to accurately model a single human brain.

To exceed that would require more!

Re:The run down says:

[LionKimbro]

1,443?

1,443 is less than 2048 (= 2**11th)

11 Moores = 22 years: x1 human brain.

22 Moores = 44 years: x2048 human brains.

33 Moores = 66 years: x4194304 human brains.

44 Moores = 88 years: x8589934592 human brains.

Re:The run down says:

[vertinox]

So, unlike what kyle90 posted, you'd actually need 1,443 (rounded up from 1,442.25) of these Blue Gene/L to accurately model a single human brain.

What about Blue Gene/P in 2006? (Which should be 1 petaflop)

What about Blue Gene/Q with 3 petaflops?

Time for Earth Simulator to make a Walmart run

[heroine]

Time for Earth Simulator to make a Walmart run and get some more Athlons to regain the top of the "supercomputer" chart.

The faster they get...

[Anonymous Coward]

The faster they make these things, the slower they sing that damn Daisy song.

Re:still en vogue?

[sgant]

When this was announced, world chess champion Gary Kasparov said "ok, no way am I playing this fricken thing"

am i the only one to notice ?

[naden]

65536 processors = 64K processors.

damn that IBM, they take geekiness to just a whole different place.

Re:am i the only one to notice ?

[ShadowFlyP]

I guess either that comment went right over everyone else's head, or I'm just imagining the reference you are trying to make: "64k should be more than enough for anyone". Bill was right: he just was talking about the wrong thing.

Re:am i the only one to notice ?

[Steve Hamlin]

65536 processors = 64K processors. damn that IBM, they take geekiness to just a whole different place.

The number of of nodes is 2^16 is not just because it is '1337, but because of the binary exponential way that the compute nodes are put together.

See how Blue Gene/L is put together.

Yes, but. . .

[02bunced]

It still takes 15 seconds to start up OpennOffice.org

Re:still en vogue?

[lightversusdark]

What do you think this machine is doing? You use this machine to calculate how far you can push your reactors to provide the energy you're talking about.

Re:Results?

[Fred_A]

And the people in ships and airplanes always love a little suprise every now and then to liven up their dull trips :)

Weather

[Voltageaav]

The main drawback to forecasting models is that it takes soo long to run all the data, so we have to cut back on the data so that we can actually see what's forecast before it happens. With this this thing running an expanded version of the GFS with 10KM resolution, we might be able to actually get it right for once. ;)

Re:Notice that performance had increased per cpu..

[mj_1903]

That probably only means that they have optimised the architecture over time as would be expected. Things like improved resource management, a slimmer kernel for each CPU, a better compiler, etc. can easily make up for that small performance gain.

gambleputtydevonausfernschpleden...

[Quadraginta]

Hmm, and yet I thought the finite speed of light was primarily an empirical fact, and perhaps secondarily a way to prevent silly violations of causality, id est to prevent everything from happening at once. And, that the Lorentz transformation was less a postulate to be applied so much as a consequence to be derived from the more fundamental notion that c should be constant in all reference frames.

Alas, I sure hope I've not been laboring under a misapprehension. I would be forced to mod myself down to -1, Doofus. Although if someone has already modded me down to -1, Offtopic or -1, Blithering, I suppose I would be modding myself more across than down -- the topology of /. modspace being kind of unclear.

I do know of people who fuss about relativistic corrections to core electron energies, but they seem a clannish, chthonic lot of Stoors, much given to muttering darkly under their caffeinated breath. I avoid 'em. Now, to me the most interesting bold-as-brass entry of relativity into ordinary (e.g. valence-shell) atomic physics is through magnetic fields. Add a pinch of vector potential to your kinetic energy operator, expand, stir, simmer, season lightly -- and, presto, fine-structure constants everywhere. Like toadstools after a good rain.

At which point the sober theorist sits back and looks quite thoughtfully at the trailing Coulumb energy term, with its implicit infinite value of c...

Uh, sorry -- what was it you were saying?

Re:Notice that performance had increased per cpu..

[Andy_R]

That probably only means that they have optimised the architecture over time

The cynic in me thinks they probably optimised the benchmark.

Meh

[Frogbert]

No wireless. Less space than a nomad. Lame.

Re:Meh

[klubar]

Lame. No USB. No firewire. No iTunes. What good is it?

many options with blue gene

[imsabbel]

The current record for blue gene wasnt running at a very high efficiency (iirc T_peak 0.7 T_max).
Plus you dont have to forget that the machince has 64K _dual core_ cpus, with one core dedicated for communication, thus the classification as 64K cpus.
There could be plenty of room for improvement by utilising this core better.

Re:Notice that performance had increased per cpu..

[Anonymous Coward]

Just because the summary says when the machine was halfway done at 138 GFlops, that definitely does not mean the other half was just putting in the other half of the processors. They probably also meant that the optimizations were not finished as well. Pure linear growth is a dream enough already. If I saw exponential growth I'd crap my pants.

Re:That's nothing ...

[catmistake]

When I was coming up, we had to use MIPS to tell how fast a computer was.

System X @ $128 Million would put out 320Gigaflops

[catmistake]

at least, right?

Re:Results?

[indifferent children]

Seem like it's just a wang dangling contest

No, this is IBM. Wang went out of business years ago.

Extra Cycles?

[Ritz_Just_Ritz]

OK, so this machine is going to simulate nuclear explosions. Based on other posts, it seems it will be able to do the same work as its predecessor in 1% of the time. Rather than run 100X the number of nuclear simulations in a given timeframe, maybe (just maybe?) the government could use our taxpayer funded supercomputer to do medical research? It seems there is now plenty of horsepower to go around unless they've been stockpiling unused simulation data for decades. Cheers,

Re:Notice that performance had increased per cpu..

[Bloater]

Linux is substantially more scaleable now than it was even just 6 months ago (not the vanilla, but quite well tested scaleability patches). This could account for the improvement. I suspect if they ran just half of it now, they'd get a little bit over half the performance (but not much over half - that is how good Linux is these days).

Re:Notice that performance had increased per cpu..

[joib]

Linux is substantially more scaleable now than it was even just 6 months ago (not the vanilla, but quite well tested scaleability patches).


Perhaps it is, but is has nothing to do with BG, since a) BG doesn't have shared memory, and each 2 cpu node (1 dual core processor) runs its own kernel and b) Linux is only used on the service nodes (the nodes handling disk IO, interactive logins, compiling etc.), not the compute nodes (where the actual action takes place).

I'm quite sure that the improvements are due to tweaking the LINPACK benchmark itself (yes, this is allowed), ESSL libraries (IBM:s version of BLAS), and improving the XL Fortran compiler.

Re:Results?

[hackstraw]

What useful utility has top sports athletes contributed to society?

Bling dude! Haven't you watched cribs?

Re:Notice that performance had increased per cpu..

[slavemowgli]

The step from 274 to 280 (really, how did you get 260 when doubling 138?) is much, much smaller than the step from 560 to 750, though. Sorry, but you're essentially talking out of your ass here. :)

HPF (Re:compiler?)

[Troy Baer]

HPF is one of those things that sounds like a great idea... until you read through a book on it and realize that they never how to talk about how to do minor things like, say, I/O. The only machine I've ever heard anybody crow about HPF performance on is the Earth Simulator, and it's not as hard to make codes perform well when you have that much memory bandwidth to throw at the problem.

Most of the codes on the Blue Gene/L at LLNL are coming from earlier ASCI systems and are most likely MPI+Fortran/C codes, possibly with OpenMP around the inner loops in some cases.

Competitive product at local dealer - when?

[smoe]

I understand this machine to be about 2^16 times as fast as my local workstation. Assuming CPU power doubles every 3 years, this would mean my grand (or great-grant) children to get such a machine as their first computer at some christmas or birthday occasion?

Today, we have the supercomputing CPU power (with respect to the 1940-70s of last century) for about every school kid. However, this has not changed society much, at least not to the positive. Or has it? Will it change society with a blue gene being available at Wal Mart in 2050? My point is not the availability of email and networks in general but the computing power that is so much more than required for normal office applications.

Re:Notice that performance had increased per cpu..

[clacour]

Sorry, but I have to disagree with your conclusion that this represents exponential growth.

The effect you speak of (doubling the number of processors giving less than double the final "power") is due to additional overhead - various processors coordinating their work with each other, deciding things like "Should I split this 2 ways or 4?" and so on - and that sort of stuff inevitably increases with the number of processors.

You can use improved algorithms, special-purpose hardware, etc, etc, to minimize this "friction", but it will always exist, and the percentage of processing that is "overhead" will inevitably climb as you increase the number of processors.

It's far more likely that either the earlier number resulted from some inefficiencies that existed then (due to it not being built as designed yet, perhaps), or there have been improvements in the algorithms or infrastructure which give greater efficiencies.

If it's the latter case, if you unplugged the 2nd half of the CPUs and made the measurement again, you'd probably get 150 GFlops or so.

Basically, you could write the equation for total power something like:

X - O - i**x, where X is the number of processors, O is the basic overhead (for doing things like I/O, for example), and c is the incremental cost of adding each processor.

To have what you describe would require that i**x be a negative number, which is like saying that you can have 10 individual conversations in less time than you can have five. Ain't gonna happen.

Power usage

[fitten]

Check out the power usage for the system and cooling.... 10MW of power (out of the 45MW the place has dedicated...)

In Soviet Russia

[Ignominious Cow Herd]

meme overuses You!

performance

[sepharious]

"That's the top end of the range IBM forecast and more than twice the previous Blue Gene/L record of 136.8 teraflops, set when only half the machine was installed." Correct me if I'm wrong, but isn't this statement a little obvious?

Skynet

[xornor]

I for one bow down to our... what's this article about?

s'more

[jasongetsdown]

I googled for a few more...

here's a closer view of a single cabinet, apparently almost completely assembled.

This one shows the overall design concept for the installation. Here again in a much sexier view

And here is a bluish picture of Gene Simmons which popped up also.

65536 ought to be...

[MrFlannel]

32 bits of processors should be enough for anyone!

64K Processors

[MarkLewis]

Nobody is EVER going to need more than 64K of RAM. Errr, I mean, processors.

Utterly off-topic grammatical congratulations...

[Radice Utente]

...to Cowboy Neal for a stroke to preserve the language. In the headline he properly spells the possessive of it without an apostrophe. In the story DIY also gets it right with "announced it's broken its own record again." English is safe for a while longer. My English degree sated, I eagerly await the mods to off-topic.

Cracking encrypted files

[WillyMF1]

What about using it for cracking encrypted files? I remember seeing a chart somewhere for how long different classes of computing (home pc, company size distributed, some old supercomputer) would take to crack some types of encrypted files. I wonder where bluejeans would stack up and how long it would take.

I looked here and it is saying that properly encrypted 128bit-key files are uncrackable.

ps. I think i saw that cracking time chart for cracking encoded zip files and it was at but the site is blocked by websense so I can't find it.

Re:Cracking encrypted files

[WillyMF1]

Damn, sorry, I should have previewed...

What about using it for cracking encrypted files? I remember seeing a chart somewhere for how long different classes of computing (home pc, company size distributed, some old supercomputer) would take to crack some types of encrypted files. I wonder where bluejeans would stack up and how long it would take.

I looked here, http://en.wikipedia.org/wiki/Brute_force_attack#Th eoretical_limits , and it is saying that properly encrypted 128bit-key files are uncrackable.

ps. I think i saw that cracking time chart for cracking encoded zip files and it was at http://www.elcomsoft.com/ but the site is blocked by websense so I can't find it.

yeah okay but

[Quadraginta]

Fair enough. I didn't say it was easy to even get basic stuff like your math libraries working well. I can believe that they've been working hard at the basic computing environment, you bet, and doing better all the time. More power to them.

But solving linear algebra efficiently is a well-studied, pretty thoroughly understood programming task. I appreciate the hard work people are doing to get it running, but I don't feel doing so is much evidence that the difficulty of good parallel algorithms for high-level tasks (e.g. docking or protein folding) is at all going away.

Also, in my experience the rate-limiting step of a high-level computation like a big molecular dynamics simulation is never the efficiency of your low-level math libraries. So, again, progress in speeding up benchmarks is not so well correlated with progress in solving actual problems.

That said, I have the feeling that this machine is going to be used largely for brute-force calculations using established algorithms and well-known existing code. I would be surprised if more than a small percentage of its cycles were dedicated to breaking ground with new and improved algorithms. But that is the work that will really pay off down the line. Frankly, I think there are few really break-through advances that have come from brute force computation. But then my training is in paper-and-pencil theory, so I'm probably prejudiced.

Re:Utterly off-topic grammatical congratulations..

[descil]

At least you don't have to be alone! I also noticed the beautifully correct apostrophe usage, and it's made my day just a little bit better. Data geek that I am, language is very important to me.

Re:Results?

[descil]

It's worse than that. It's the world's most expensive computer, paid for by a company who isn't spending their own money. Whose money are they spending? Why, yours of course! (If you live in the US.)

It's not about the pissing contest, and they're definitely making money off of US Government contracting - which is a particularly nice way to waste money, since it's someone else's money, it makes you tons of extra money, and you don't have to worry about competition! Woohoo, go US!

*sigh*

Doesn't hold a candle to me

[LilGuy]

Lets see it beat my 500 terafaps record set when the videos of that chick with the perfect ass came out.

Some tasks are inherently easy to split up

[Dog135]

There are certain tasks that are easy to split into individual threads. Just about anything graphical deals with single pixels at a time, and complex 3D simulations deal with voxels. A simple 100x100x100 simulation would use 1,000,000 voxels. Each processor would handle a handful of voxels, and communicate with computers simulating joining voxel groups.

Just some more stats:
for a 1024^3 voxel cube, with 64K processors, each processor deals with 16K voxels.
16K^3 voxel cube: 64M voxels per processor. (A few Gigs of ram each should be enough)

Inaccurate and irrelevant

[Dog135]

The article says that the brain fires at 1000 times a second, but practice shows that we can only process data at 30 pieces of data per second. Plus they forgot that only 10% of our brain is in use at any one time. To taking that into account, you can multiply their estimate by .003.

120,000,000 cpus * .003 = 360,000 cpus. (only 5 times the size of the one in the original article, so it could simulate a human brain at 20% speed)

But all of that is irrelevant anyways. CPU power isn't the problem. It's knowledge of how the brain works. A honey bee's brain only has 960,000 neurons. Yet they search for food, remember where their home is, they fly, walk, eat, reproduce, and perhaps most importantly, they communicate. (through dance)

960,000 neurons, at 30 "frames per second" and 100% brain use is only 28 million values per second, and less then 4 Megs of RAM. Even if you go with their number of 1000 "frames per second", it's still only 960 million values per second. Well within the processing power of a single computer. That's assuming a neuron can be represented by a single 32 bit integer.

Has anyone simulated a bee's brain yet? No? Well why not? They have complex behavior, they're even able to communicate about how much food they found, and directions on how to get there. (including distances!) No, the real problem is that no one know how the brain actually works. This IBM cluster can simulate an entire hive of honey bees, but they just don't know how. Until then, they'll never simulate a human brain.

Code Bloat 2050

[Dog135]

My point is not the availability of email and networks in general but the computing power that is so much more than required for normal office applications.

I still have my MS Word 4.0 for my Mac Plus. It was on a single sided disk with enough room to store documents. On my plus, at about 8Mhz, Word opened about as fast as the 2000 suite opens on my work computer, which is about 1Ghz. Trust me, office applications will bloat in direct proportion to the speed of your computer.

In 2050, they'll have all these new features though! Like spell and grammar checking all the documents on your HD every time you hit a key. (all loaded and held in RAM, of course) Plus, lots and lots of fancy, animated 3D graphics and dolby 5.1 surround sound for the sound effects. Every time you hit a key, you'll hear a sonar ping wiz past you in your office. (plus a floating, 3D, ray-traced version of the letter you hit rotating at the center of your screen) Even when you turn off the options, they'll still be generated by the CPU, just in case you turn them back on in the middle of an animation or sound playing.

Don't say it won't happen.

Does it run Websphere?

[matresstester]

Can I get my J2EE webapp to run on it? Does it automatically overcome my bad code? Does it have enough memmory to tolerate the IBM xsl memmory leak?

The real story

[Blitzenn]

The real story is that Blue Gene was supposed to run Windows Vista code. At the time, MS supplied an alpha set of code to trial on the machine. A massive concurrent multiprocessor buffer overrun caused a blackhole to develop many miles above the earth. Unfortunately this was the exact same time the that Space Shuttle Columbia was passing through the same space. It in turn caused the shielding in the wing to be ripped away and.. you know the rest.

Entire issue of Journal of R&D devoted to BG/L

[Bill Privatus]

See Volume 49, Number 2/3, 2005.

Re:Notice that performance had increased per cpu..

[afidel]

By optimizing for the benchmark the have optimized for a specific class of problem that the cluster may need to do in it's "real" job, ergo they have made better at doing what is supposed to do.

LINPACK is a collection of Fortran subroutines that analyze and solve linear equations and linear least-squares problems.