25th TOP500 List Released

Chris Vaughan writes "The 25th edition of the TOP500 list of the world's fastest supercomputers was released today (June 22, 2005) at the 20th International Supercomputing Conference (ISC2005) in Heidelberg Germany. The No. 1 position was again claimed by the previously mentioned BlueGene/L System. At present, IBM and Hewlett-Packard sell the bulk of systems at all performance levels of the TOP500. The U.S is clearly the leading consumer of HPC systems with 294 of the 500 systems installed there (up from 267 six months ago)."

Comments

Obvious Link?

[yellowbkpk]

The list can be found here:
http://www.top500.org/lists/plists.php?Y=2005&M=06

Re:Obvious Link?

[neosake]

Wtf?! did i read it correctly?!
Processors: 65536
That's more than 10 times the earth simulator!

Re:Obvious Link?

[Anonymous Coward]

Looking at the number of POWER systems on the top500 really shows how braindead a decision by Apple to drop their CPUs at this time. Anyone think it's likely apple will still keep powermacs available after the intel switch, even at a higher price for places like System X?

Re:Obvious Link?

[Anonymous Coward]

Japanese and American supercomputer technologies are different. You want to compare the performance scores instead. The Rmax values: 136800/35860 = 3.8

Re:Obvious Link?

[spauldo]

POWER != PowerPC.

PowerPC is _based_ on POWER. The G5 is basically a modified and scaled down POWER 4 chip.

Apple's got other concerns rather than just raw computing power, and they don't need the features that allow you to have more than 4 or so processors in one system. POWER itself isn't designed for small applications - engineering workstations is about as low end as it gets.

It does suck though. PPC's a nice platform.

Re:Obvious Link?

[syphax]

Look at the trends, though. Power's Top 500 system share peaked in 2000. For better or worse, Intel is on the rise...

Re:Obvious Link?

[Anonymous Coward]

How does the number of Power systems (10.4%) and PowerPC systems (5.0%) show that choosing Intel (66.6%) is a braindead decision?

Re:Obvious Link?

[quarter]

What? If you are going to use number of systems as your metric you might want to check it a little closer : 5% for PowerPC vs. 66% for Intel makes it look like a great decision.

Re:Obvious Link?

[joeljkp]

So who here has access to one of these monsters?

I'm logged into #367 right now.

Re:Obvious Link?

[Anonymous Coward]

Well I beat you there sister!! I am logged on to #38

Re:Obvious Link?

[Doc Ruby]

What are you running? And what do you hope to learn? Please don't tell me it's pr0n, unless it's superpr0n ;).

Re:Obvious Link?

[50000BTU_barbecue]

Speaking of ignorant, have you proof-read your sig lately?

"those that would give up essential liberty for temporal safety deserve neither liberty nor safety."

I happen to think temporal safety is very important, thankyousoverymuch. All that causality and stuff, ya know?

Re:Obvious Link?

[joeljkp]

Molecular dynamics with Discover

Re:Obvious Link?

[Chmarr]

One word... laptops.

IBM do not have a PowerPC (or POWER) solution for laptops.

Why do people keep ignoring this very, very important facet?

Re:Obvious Link?

[Doc Ruby]

Ah, computational chemistry. You don't know of any Linux visualization software for supercomputed datasets that let the user play in Feynman's QED model, do you?

Re:Obvious Link?

[DrFrob]

Beat you there. #28.

Molecular dynamics simulations.

Re:Obvious Link?

[Anonymous Coward]

http://www.techweb.com/wire/networking/164901888 Story with quotes from Top 500 list creator

Re:Obvious Link?

[fimbulvetr]

If you find one, lemme know, please.

Re:Obvious Link?

[devinoni]

BlueGene/L is also much smaller than Earth Simulator. At 65536 processers you get 32 cabinets (2048 per). While Earth Simulator is 320 cabinets for the CPUs alone, not including the 65 cabinets for the interconnects. Construction of BlueGene/L is not complete it will have 131072 processors when it is fully completed.

Earth Simulator Facts
BlueGene/L Facts

Re:Obvious Link?

[Anonymous Coward]

No kidding. It was like 'Holy freaking crap, Batman!"

Re:Obvious Link?

[Single GNU Theory]

I work in Oak Ridge, TN, so I'm at least near to four Top 500 systems. That's as good as it gets for me.

Re:Obvious Link?

[pomo monster]

Oh for God's sake. You say doctors and lawyers get more respect than teachers--doctors, fine, but lawyers get made fun of too, and brutally at that. I think it has something to do with the idea that "anyone could do that job," even though everyone knows it takes a special dedication to be a lawyer, or, for that matter, a teacher. The lawyers I know can take the ribbing. So can the teachers. Because really, in the end, it's just a joke.

Re:Obvious Link?

[Danta]

Just as interesting, the Top 500 list with MFlop/$ information, i.e. which system gives the most bang for the buck.

Interestingly Apple's G5 based supercomputers are a lot cheaper than the rest. But there is also one very cheap Xeon-based system.

Re:Obvious Link?

[InvalidError]

A PowerPC is just a single core... a POWER4 is a multi-chip module featuring four qual-core CPUs with integrated RAM controller, insane internal and external interconnects and 36MB L3 cache. The POWER4 MCM itself has >5000 pins and peak power dissipation around 400W... not quite in the same league as off-the-shelf PowerPC cores.

The POWER4's PPC is different from the Cell's PPC which is different from G4/G5's PPC and the PPC that will end up into the XBox360... IBM likes to customize the PPC to meet big accounts' requirements.

Re:Obvious Link?

[spauldo]

Four points:

A) I did tech support for two years, then did it on a rotating schedule while in the military, then ran a computer shop where we did it all the time. I still do for a couple of clients.

B) Becoming a teacher is one of my plans for after I get my master's degree. While in Japan, I spent maybe a third of my time training new troops who were new to network administration. I deal with teachers every day at school, and get along great with the vast majority of them.

C) There's a lot of examples of people who, in times of economic downturn, become teachers. There's even more examples of people who teach training classes who barely know the material and have almost no experience in the field. These are the people referred to in my sig.

D) You take this way, way, way too seriously. Get some ex-lax.

Links are Fun

[TPIRman]

And here's a link to the actual list. Also interesting is the historical chart of the TOP500 by manufacturer, which tells a story in itself -- the decline of Cray and rise of IBM and Hitachi, for one.

Re:Links are Fun

[BasilBrush]

ITYM IBM and HP.

Re:Links are Fun

[Elrond]

I believe you mean HP and not Hitachi.

Re:Links are Fun

[TPIRman]

Quite right -- my apologies.

Re:Links are Fun

[markbo]

...and the link to the Top500 list sorted by operating system -- which tells the story of the rise in Linux.

Link

[avij]

How about a link to the new list? Yes, it can be found from the TOP500 website, but that link was missing as well.. clicky clicky.

The full list

[TheAngryArmadillo]

For those that thought they might like an actual link to said list:

Here ya go.

Already!

[SLASHAttitude]

One person post a link and it is already slashdoted! Maybe they should try running there web server on something other then a toaster.

Re:Already!

[SLASHAttitude]

T3, Ass!

Re:Already!

[SLASHAttitude]

Damn, I knew it had to be somebody doing that. If you stop maybe I can get to the photos of your mom.

Re:Already!

[Anonymous Coward]

Deal. You know the IP.

LINK?

[ebynum]

Where's the link?

Re:LINK?

[Leroy_Brown242]

http://hardware.slashdot.org/comments.pl?sid=15357 0&cid=12882187

read, before asking.

So where is the list?

[SeanTobin]

You'd think that it would be a good idea to actually link to the html list, or the xml list, or the pretty charts.

The press release is interesting too.

Re:So where is the list?

[bhsx]

The interesting thing about those charts are the over 50% unspecified on both. How much of that is clasified government work, and what is it crunching?

Re:So where is the list?

[Kent Recal]

Maybe just people who didn't bother to fill in the "purpose"-field when they rented some CPU time? I somehow doubt classified stuff would appear in these charts.

Re:So where is the list?

[Anonymous Coward]

I just browsed through the list and the thing that strikes me, above everything else, is the sheer number of 2005 BlueGene machines (5 in the top 10!) that are on the list. And not only that, but the fact that identical BlueGene machines, purchased by different organizations, presumably for different purposes, achieve *exactly* the same benchmark. See for instance numbers 8-9, 21-23 and 58-64.

To me, this means that IBM has found a way to more-or-less mass-produce supercomputers. And that's a big break from the past: The type of supercomputers that traditionally made it to the Top 500 were always one-off machines, built by a company as a special project for a special customer. Those days seem to be over now.

Derived Moore's Law

[OlivierB]

It would be great if we could verify Moore's law through some simple stats using the histrical data from this Top500 list.
-For example:How many years did it take for Number ones on average to be dropped off the 500 list?

- How many years after the list was published did it take personal computers tu make it in the 500list? To make it to the number 1 spot?

- How many transistors did these computers have? Did it verify Moore's law?

- Are we getting more TFLOPS per watt now? Per transistor?
etc..

Re:Derived Moore's Law

[0kComputer]

If I recall correctly, Moore's law only applies to Silicon Based machines. Supercomputers typically arent silicon based.

Re:Derived Moore's Law

[Anonymous Coward]

What are you smoking? Almost everything on the top-500 is essentially a cluster using commodity silicon-based chips.

Yes, at one point, supercomputers used higher-performance "unusual" semiconductors. But it's just too expensive. It's all about bang for your buck now, and that clearly favors silicon chips.

Re:Derived Moore's Law

[ndansmith]

Maybe we should measure computer performance in more practical terms. Maybe it should be a function of input and output. For how much work the user puts in, how much work does the computer put out? That is the real point of computing, having a machine do work for us, so perhaps it would help to measure power in more concrete terms.

Re:Derived Moore's Law

[eln]

That is the real point of computing, having a machine do work for us,

Dude, are you high? Everyone knows the real point of computing is playing games and viewing pornography.

Re:Derived Moore's Law

[LordIvan]

Yeah, a *real* supercomputer uses vacuum tubes and steam valves.

Re:Derived Moore's Law

[Short Circuit]

Good point. It's just not the same if you're not looking at the raw MPEG binary data.

Re:Derived Moore's Law

[slavemowgli]

How do you measure the "output" of computers in terms that actually make sense from a practical perspective, though?

That's opening a whole can of worms really. A computer is just a tool, and this kind of measurement is simply not possible. As another example, take a screwdriver - pretty much anyone would agree that it's a tool that makes life easier, but you simply can't measure the output of a screwdriver.

Computers are similar, and the reason for that is that while computers are machines that process data and thus have a "natural" input and output if you view them from a close perspective, if you step back to take a look at the bigger picture, it's exactly that natural definition that becomes irrelevant. This is implied in what you write, too: a computer that is used today may have a much higher raw computing power than one from 20 years ago (FWIW, it may well be a thousand times faster, if not more), but that kind of figure is meaningless unless you are doing pure number crunching.

All in all, I'd actually go so far as to say that if you really found a way to measure the amount of work a tool saves you in an objective and meaningful way *without* actually doing the same work twice (once with the tool, once without), then you'd likely get a Nobel prize for economics for it - pretty much all of economics seems to be centering around attempts to get good approximations for things like this, and the best (that is, most successful) managers are often those who simply have the best instinct in this regard (or the ones who simply had the most luck, if you subscribe to a more Dilbertesque point of view :)).

Re:Derived Moore's Law

[DerekLyons]

It would be great if we could verify Moore's law through some simple stats using the histrical data from this Top500 list.

That would be nearly impossible - Moore's Law applies to individual IC chips only, and the various machines contain wildly verying numbers of chips/cpu's/etc...

Re:Derived Moore's Law

[danila]

In fact, the historical performance of supercomputers is one of the strongest proofs that the Moore's Law will continue unabated forever (*)

http://www.top500.org/lists/2004/11/perfdevel-spec ial/projected.php

The total performance of the 500 fastest supercomputers has been increasing almost perfectly exponentially from 1993. The graph is very impressive.

* Insert standard disclaimer about forward-looking statements here.

AlphaServer holding strong at 12th

[NRAdude]

With a 5 year-old design, Alpha architecture is still holding strong at 12th position;

ASCI Q
ASCI Q - AlphaServer SC45, 1.25 GHz

That'll be the hardware on my list to buy when they decide to move (downgrade) onto crippleware Intel DRM architectures.

Re:AlphaServer holding strong at 12th

[diegocgteleline.es]

Wrong, Alpha is still fourth. With half of the processors of the 3rd competitor (5120 vs 10000) it achieves more than half of its "R peak". I guess there're lot of factors (interconnexion use, etc) here but it'd be interesting to see a performance/number of CPUs chart

Re:AlphaServer holding strong at 12th

[joib]

Wrong. The forth computer you linked to is the earth simulator, which uses NEC SX-6 vector chips (which have nothing to do with the alpha). That's why its processors are so fast compared to scalar processors (and insanely expensive too, for that matter, which is why most supercomputers use commodity processors).

Incomplete ranking

[Anonymous Coward]

Why is my beowulf of Mac Mini's not in the list?

Re:Incomplete ranking

[richdun]

Because my beowulf of Linux-running Nintendo DS machines from the other day beat your Mac Minis.

Re:Incomplete ranking

[Whafro]

Why is my beowulf of Mac Mini's not in the list?

...Of course, it was because you used an apostrophe to pluralize "Minis." That's one of TOP500's pet peeves.

Re:Incomplete ranking

[Anonymous Coward]

Because Apple hasn't made enough of the under powered Minis so that all of them together will even make it to #500.

And calling your two Minis on a x-over cable a Beowulf Cluster is insulting (to those of us with three and a switch).

Imagine...

[Mathiasdm]

Imagine a Beow...

Hmm, nevermind.

Re:Imagine...

[Anonymous Coward]

Imagine people not using the "Imagine a Beowulf cluster of -" joke . . .

Hmm, nevermind

No PS3?

[0kComputer]

How can this be? I thought it was running at 2+ Terraflops. Didn't anyone watch E3?

Re:No PS3?

[elrous0]

You see, the PS3 runs so fast that it actually broke the light barrier, travelled back in time, and made last year's list.

-Eric

Re:No PS3?

[Anonymous Coward]

It was just a marketing way to make people like yourself believe it's more powerful than it really is.

Re:No PS3?

[djbentle]

Duh! The PS3 isn't out yet. Don't worry it will be there next year. Up near the top!

Re:No PS3?

[0kComputer]

Im guessing my attempt at humor was lost, or maybe wasn't obvious enough :\

Re:No PS3?

[sharkey]

I thought it was running at 2+ Terraflops.

Because it was obviously a lie. There is no system at this time that can do 2+ Planet Earths per second.

Re:No PS3?

[slavemowgli]

OK, I do realize that you're just joking, but it's actually an interesting question, so here are some thoughts:

1. The PS3 isn't out yet.
2. The top500 lists does NOT go and look for systems that should/could/might be included - rather, you have to submit benchmark results yourself. I assume they do check them, of course, but as long as you don't approach them about it, you could have the fastest system in the world, and you'd still not get listed if you didn't submit benchmark results (incidentally, this means that the list is a priori worthless for assessing the actual computing power wielded by different governments, intelligence organizations etc., and that there's no reason to be paranoid about what the systems are really used for - what you should be paranoid about are the systems that aren't even listed).
3. The top500 list is based on LINPACK performance, and it's no a priori clear that the PS3 would achieve an outstanding performance there. Now, maybe it would, but I assume it's similar as with GPUs - it's optimized for high performance in a specific area, and that area isn't necessarily solving dense systems of linear equations (which is what LINPACK is all about).

That being said, it'll be interesting to see whether there'll be Cell-based machines (other than the PS3) in the list in the future - IBM likely has at least a few aces up their sleeves, and the Linux port to the Cell architecture is well underway (check lkml - an updated set of patches was just submitted for inclusion yesterday).

Re:No PS3?

[The One and Only]

Actually that would be an interesting measure of performance for future computers, to see how quickly they could fully render the planet Earth like the Earth Simulator. 2 planet Earths a second would mean that for every second two full renderings could be completed, which seems like a pretty good rate (real-time for many practical purposes).

Re:No PS3?

[Hoser McMoose]

It's also worth noting that the PS3 and the Cell processor are strongly targetting single-precision floating point operation, and that is where the 2 TFlop number comes from. The Linpack test used for the Top500 list requires double-precision floating point operations. I have yet to see numbers for the Cell processor in double-precision floating point, though I would expect that it will be significantly less than half of it's single-precision performance.

FWIW this is the same reason why SSE2 can be used in Intel and AMD's newest chips, but Altivec and the original version of SSE don't cut it.

Re:No PS3?

[Anonymous Coward]

But you would have to determine just how much detail would be considered a full rendering of Earth. Is it just the view from space, or do you render it right down to the trees? Or do you go even further and render the grass properly or there is always the furtherest extreme, rendering it right down to the atomic/subatomic level?
Which is considered a full rendering of Earth?

Re:I looked at one of the links

[ZackSchil]

Orisinal.com is a site that makes flash and shockwave games. Damn good ones too. I'd expect nothing less.

Position #501

[Nom du Keyboard]

And at position #501, OSX running on an Intel processor. Hey, Steve promised it would be fast.

Re:Position #501

[pHatidic]

You must be new around here. OS X on Intel isn't faster, it's snappier.

Re:Position #501

[operagost]

And lickable.

Re:Position #501

[daviq]

Didn't you see the snappier keynote on intel at wwdc 2k5?

Re:Position #501

[kargis]

You must be new around here. OS X on Intel isn't faster, it's snappier.

Perhaps faster, perhaps not, perhaps snappier, perhaps not. What it will be is PORTABLE without causing 2nd degree burns as a G5 powerbook would. The problem with the PPC970 is that it's just too darn hot -- witness the water cooled production machines (dual 2.7 ghz G5). You can't do that with a notebook and have it be meaningfully portable. This is probably the biggest reason they had to ditch PPC, along with the lack of speed improvements in the G5.

Kargis

A Simple Test.

[OctoberSky]

But how would they fair against being slashdotted?

Link us to thier HDDs please!

You know its coming

[dankasfuk]

//insert generic pr0n joke here

Testing....

[zappepcs]

Are any of these supercomputers rated for testing BT bandwidth limits.... or making a website /. proof?

Come on Canada!

[matrix mechanic]

The biggest one in Canada, Westgrid's Glacier, adds 600 nodes, and still slips from 54 to 83. This obviously calls for one up man ship. Oh Canada.

Re:Come on Canada!

[Spez]

Thaths so untrue!! University of Sherbrooke is in 40th position. I've seen the system (named Mamooth) and its quite cool :) 1152 Xeon 3.6ghz

Re:Come on Canada!

[RandomBitFlipper]

Yeah - i noticed University of Sherbrooke a couple times on the list... surprised me.

I wonder what they're crunching in the Eastern Townships... calculating the costs of separation from Canada? :-p

Re:Come on Canada!

[Spez]

University of Sherbrooke got the 1st and the 3rd most powerfull computers in Canada. A lot of research in there :) (AI, neural nets, etc)

Re:Come on Canada!

[Anonymous Coward]

Nah, what would be surprising is if ETS is on there... Those guys can barely tie their shoelaces, from what I've seen there...

Re:Come on Canada!

[pkhuong]

Probably figuring how to make Johnny Charest win the next elections...

Re:Come on Canada!

[Anonymous Coward]

They have computers in Canada? I thought they barely had electricity there... Abaci made out of ice run by polar bears don't exactly count.

Re:Come on Canada!

[sapbasisnerd]

i noticed University of Sherbrooke a couple times on the list... surprised me

Quebec has something like 25% of the population of Canada and produces 48% of the engineering and science graduates.

There are plenty effed up things about Le Belle Province but their approach to technology education isn't one of them.

surprsing to me

[udderly]

What's surprising to me is that Cray used to be synonymous with supercomputers and they now have comparatively few entries.

Re:surprsing to me

[Anonymous Coward]

Don't forget that Cray was acquired by SGI

Cray is #11, SGI is #3 and #15

Cray was really only on top when they made the FIRST supercomputers. Hard to be beat when you're the only player in the game.

Re:surprsing to me

[Tenebrious1]

What's surprising to me is that Cray used to be synonymous with supercomputers and they now have comparatively few entries.

Why is that suprising in any way? At one time, Ford was synonymous with cars, but today have news of Ford laying off managers. IBM used to be synonymous with the desktop PC, but with the sale of their laptop division are now completely out of the market. Sony Walkman was synonymous with portable music, but now everyone has an iPod.

Cray is just another company that had a great product for a while, but couldn't keep innovating and couldn't keep up when the competition joined the market. Nothing at all suprising about it, it happens all the time.

Re:surprsing to me

[zagava]

Cray was acquired by SGI, they took the technology to put in thier altix line and then sold them off.

Re:surprsing to me

[Anonymous Coward]

You don't understand what has happened here.

Cray makes some of the fastest machines in the world, right behind IBM's Blue Gene.

The reason they are not ranked higher is that the benchmark code (linpac) only measures raw computing speed, not things like network latency, network bandwidth or the memory heirarchy, which is where the Cray machines excel.

There are people addressing this with the creation of the HPC Challenge benchmarks.

Re:surprsing to me

[afidel]

While Linpac may be a bit more parallel then some supercomputer apps it's comprised of real workloads. This means that for some cross-section of the supercomputer market Linpac is a good measure of a systems performance. And of course the people who are given millions to buy a supercomputer generally know the differences between different architectures and why they would want one system over another.

Re:surprsing to me

[k98sven]

What's surprising to me is that Cray used to be synonymous with supercomputers and they now have comparatively few entries.

Cray still makes some of the fastest supercomputers around. They do not, however, make supercomputing clusters, which this list includes.

So you're comparing rather different things. And it's an important difference since not all computing tasks can be parallelized.

Re:surprsing to me

[oncebitten]

checking for alloca ... yes
checking if alloca needs Cray hooks ... what's a Cray?

Re:surprsing to me

[rhennigan]

Nothing at all suprising about it, it happens all the time.

Would be nice if it happened in the software world...

Where's the Sony PS3?

[Spy der Mann]

I've been cheated! :(

Re:Where's the Sony PS3?

[vikstar]

The graphics constitutes 1.8 Terraflops of the ps3's 2 Terraflops, hardly what can be called useful computation. I have a pencil on my desk that can do easily over 200 Exaflops by maticulous "calculation" of how photons that hit it should be absorbed or reflected to create the stunning diffuse, specular and anisotropic lighting that I am observing.

Choosy Supercomputers prefer *nix

[DebianDog]

Sorry I did not see an anti-M$ post yet ;)

Re:Choosy Supercomputers prefer *nix

[Nom du Keyboard]

Sorry I did not see an anti-M$ post yet ;)

Until now.

Re:Choosy Supercomputers prefer *nix

[DebianDog]

Well I figured since Apple managed a #14 slot M$ could at least "show up" ;-)

Re:Choosy Supercomputers prefer *nix

[Potader]

... still installing service packs and patches.

Re:Choosy Supercomputers prefer *nix

[AJWM]

Well, M$ doesn't make hardware.

Actually, come to think of it they do. Where's the Beowulf cluster of XBoxes?

Re:Choosy Supercomputers prefer *nix

[mangu]

It seems that M$ doesn't enter into this picture. But it's interesting to see how Linux is grownig.

TOP500 List???

[Anonymous Coward]

Some TOP500 list, where's the warez, Pr0n, and Roms?

FTFA

[Doc Ruby]

Slashdot has finally evolved to its ultimate form. A story "summary" that links only to a previous Slashdot article. If only this post consisted of only "FR1ST P0ST!", I could retire my userID.

Re:FTFA

[Anonymous Coward]

Slashdot's final ultimate form will be achieved if this story is duplicated with a slightly different summary and the same links, i.e. no link to the actual site. :)

Re:FTFA

[Doc Ruby]

Moderation 0
20% Flamebait
30% Funny
30% Overrated

OK - with the "Flamebait" TrollMod, this story is complete. But that "Funny" is keeping my userID in action.

All this computing power

[Dancin_Santa]

Despite all this computing power, computers still can't think like humans. They can perform calculations faster, but can't perform optimized heuristics or even form optimized heuristics like humans.

Computers are pretty dumb. Humans are amazingly smart.

Re:All this computing power

[bnavarro]

Personally, I don't think that Human brains are binary based, logic gate controlled computation machines, and this difference accounts for why we have so much diffuclty with developing strong AI on them.

I do believe, however, that we will eventually "crack the code" to the fundamental archetecture of our brains, and once we do that, we will re-design our computers accordingly, and finally achieve strong AI.

I also believe, that our currently architected computers will play a key role in assisting us with cracking this code.

Re:All this computing power

[agent dero]

Computers are pretty dumb. Humans are amazingly smart.

I live in Texas, and..well, beg to differ :-P

Re:All this computing power

[Bimo_Dude]

I don't think that Human brains are binary based, logic gate controlled computation machines, and this difference accounts for why we have so much diffuclty with developing strong AI on them.

Every day I see evidence that we have so much difficulty developing strong AI on the Human brain.

Sorry... I couldn't resist ;)

Re:All this computing power

[Waffle Iron]

Despite all this computing power, computers still can't think like humans. They can perform calculations faster, but can't perform optimized heuristics or even form optimized heuristics like humans.

That's probably because brains use a completely different architecture than digital computers. Neurons connect in a highly parallel fashion, with trillions simultaneous of connections arranged in 3D directly between various parts of the brain. Even with the 1000000X speed advantage of computer logic, the number of permutations of neuron connections compared with the serial nature of computer buses allows the brain to outpower computers on many real-world problems.

Because they are full of narrow bottelneck data paths, computers rely heavily on locality of reference and precomputed indices to do anything efficiently. A brain, with a storage architecture approaching fully associative memory, can instantly compare any input against a lifetime of experiences with no need for predefined indices. It is somehow able to use high-level concepts as access keys as well, in contrast to the binary numbers that computers must use to address storage.

The result of all of this is that for many tasks like navigation in the real world, a cockroach brain compares favorably to the most powerful current digital computers.

Re:All this computing power

[sbackholm]

One day we may re-design our computers to achieve strong AI for Man, but for Woman--this is at least another hundred years out.

Or is that one of those NP-complete problems?

Re:All this computing power

[Anonymous Coward]

Personally, I don't know how Human brains work, but instead of doing any basic research I'm going to spout my own pet hypothesis and claim that it solves everything.

Everything can be represented in binary. If you can't understand that part, there's no need to continue explaining; you are incapable of being taught.

Re:All this computing power

[Vitriol Angst]

I don't think we will get current computers (binary) to simulate the human brain. It will take a new platform that allows for simultaneous and multi-spectrum data (like optical computers) to work. The human brain stores information 3 dimensionally--meaning in a network of neurons and not discretely, like one neuron holding a picture of the family dog. The human brain is also horrible at storing perfect data; like a series of 10 numbers, whereas any computer can easily store 10 million numbers and not forget.

In an optical computer, I foresee a Green pulse of 5 magnitude, a Blue of 7 magnitude and a Red of 2 magnitude and other frequencies measured by direction, frequency and hue all being processed simultaneously. In our brains, we continuously have simultaneous thoughts all the time -- we only can focus on one thought at a time. The one thought on our mind is decided by a region at the left frontal lobe (sensitive to certain drugs used for hyperactivity). Priority is given by strength of signal. So you could be looking at a sunset and not notice a mosquito bite until later -- but a bee would make you jump. Binary cannot process these values with "meaning".

Computers are absolutists while brains deal with relativity.

I'm confused

[Locke2005]

Why do some machines acheive close to 90% of their theoretical max (Rpeak), while others get only about 50%? Is it communications bandwidth that makes all the difference?

Re:I'm confused

[CardiganKiller]

It all depends on the system architecture and the type of problem being solved. Certain problems will adhere better to certain architectures and thus allow for a smaller gap between the theoretical and actual performance. The gaps can also be inherent in the architecture itself (e.g. communications bandwidth like you said).

Re:I'm confused

[fitten]

Yup... and the benchmark used for Top500 is pretty simple and is only really relevant to certain algorithms. It's a pretty simple and forgiving benchmark that even clusters can do well on.

Re:I'm confused

[Anonymous Coward]

It's "balance" -- peak is just what the FPU can do.

Inter-node communication is a big part of it. (Have a look at % peak obtained by the same CPU over differnet interconnects.)

Re:I'm confused

[Locke2005]

Uh, for purposes of Top 500 List classification, they are all solving the same problem -- the High Performance Linpack benchmark (solving a system of simultaneous equations via Gaussian elimination with array pivoting). Granted, I beleive some variance in the size of the arrays is allowed, giving more massively parallel machines that can handle larger arrays an advantage.

Re:I'm confused

[Anonymous Coward]

A combination of bandwidth and latency can do that.

In reality (unless your reality is running linpac) most codes only get 10-20% of peak.

Re:I'm confused

[Anonymous Coward]

Network bandwidth or latency is certainly a common reason for low latency. On my cluster (which is in the top 50), the reason for ~50% efficency is that we only have 500MB of ram for each of the ~2000 processors.

Back when we had ~1000 processors, our efficency was over 70%. If we had 2G of ram for each processor today, we'd beat 70% again.

BlueGene domination

[javaxman]

BlueGene/L max linpack: 136800
Earth Simulator ( #3 on the list ) : 51870

The #1 linpack score is well over twice the #3 linpack score ?!?

That fact combined with the large number of IBM-based systems on the to 100 list really makes it look like IBM is dominating this sector of the market.

You know what data is always missing from this list that we'd all like to see ? The cost of the systems. Although, I suppose if you're looking at building the most powerful computer system on the planet, cost might not be your first consideration...

Re:BlueGene domination

[reverseengineer]

First off, #3 on the list appears to be NASA's "Columbia" supercomputer (Earth Simulator is #4). Your point is worth commenting on, though. IBM holds both of the top 2 spots with their amazing Blue Gene systems, but it appears that some of the systems down the list get more GFlops per processor- note that Lawrence Livermore's Blue Gene uses 65536 procs to deliver an Rmax of 136800, while NEC's Earth Simulator gets 35860 from 5120 procs. That's roughly 7GFlops per processor for the E.S. versus 2 and change for the Blue Gene machines (and I mean very roughly- I understand that things like interconnect speeds and instruction sets can be just as important as processor speed). Similarly, Blue Gene/L gets less than three times the Rmax of Columbia using over six times the procs.

That's no knock on the Blue Gene systems; fastest is fastest, after all- it's just interesting that there are a wide variety of ways, ranging from clusters of commodity hardware to bespoke vector processor systems, to build a top supercomputer.

Re:BlueGene domination

[Anonymous Coward]

That's why the number 5 showing of the JS-20 running linux was interesting. This is the lower-end IBM version of Apple's G5.

Supposedly there is a US Navy cluster of these as well. Apparently over 10% of Apple G5 hardware is running linux for computation, so I'm not surprised to see some of these clusters hit the list.

Re:BlueGene domination

[Retric]

What I want to see is the list of ongoing costs. I see a tun of PIV's on the list but I wonder how much more it costs to use those PIV's say Athlon 64's ect. Every extra watt adds up.

At ~10c per kilowatt hour * 95% uptime = 0.83$ / watt per CPU / year.
So ~100w CPU * 65,000 CPU's * 3 years * 2 (AC costs) = around 3.2 mill on energy costs.

Not that I expect them to dump the system in 3 years but the energy costs to run a system like that stop being worth it after a few years and then it's time to upgrade.

Re:BlueGene domination

[afidel]

When the Earth Simulator first hit the list is was as fast as the #2-#5 machines combined and was almost five times faster than the number two entry.

Re:BlueGene domination

[devinoni]

Remember the goal of BlueGene is to build very dense systems. Not only do you have to factor in the costs of the system, but you have the costs of the facilities. This includes costs of construction or renovation of the facilities to handle the power and cooling requirements of these behemoths. BlueGene/L in it's current incarnation is using 32 cabinets for it's processors. While Earth Simulator is comprised of 320 cabinets for the CPUs (an additional 65 for interconnects).

Re:BlueGene domination

[stevesliva]

You're just about one order of magnitude too high on the power consumption, if you're referring to BlueGene. 10-15W peak power per dual-core "node."

Re:BlueGene domination

[k98sven]

You know what data is always missing from this list that we'd all like to see ? The cost of the systems.

Well, Lenngren (#65) is a few months old, and costed (at the current exchange rate) about $250,000.

So, not as bad as one would think perhaps, it's commodity hardware after all, except for the infiniband networking hardware, really.

Re:BlueGene domination

[evilviper]

That fact combined with the large number of IBM-based systems on the to 100 list really makes it look like IBM is dominating this sector of the market.

It certainly is. The sad thing is how it came about.

Up until about last year, HP/Compaq/DEC made up the majority of the list. Compaq slowed Alpha development to a crawl at that point because they were ready to switch to Intel Itanium. HP had some good high-end machines on the list too, and they killed off their propritary processor lines because they were ready to switch to Intel Itanium. (SGI fell for Intel's same trick, too)

The giant vaccume in the supercomputer market was purely due to Intel managing to trick every supercomputer company into betting the farm on Itanium, and never delivering on their wild claims.

IBM and Sun were the only major players who didn't get suckered in by Intel. Sun isn't exactly known for having fast (or cheap) machines, so they weren't in as good of a position to take over. IBM, though, was in the perfect spot. There was practically no competition, and after about a year with nobody else jumping in to take-over, they developed faster and low-power chips, and started building systems that creamed everyone else.

HP/Compaq/DEC still have some systems on the TOP100, but they are mostly older systems from back in their better days. It's really quite sad when all the players lose, not because they couldn't keep up with the competition, but just make a stupid decision to quit all together.

Re:BlueGene domination

[Retric]

Wow that's low. There is a tradeoff between cheep limiting yourself to "perfectly" parallel code so 15W per node seems extremely low but I guess they know what their doing.

Anyway, do you know of any chart that keeps track of a top 500-type list that includes things like ongoing costs and performance on highly interconnected simulations?

PS: Thanks for that info where did you find that out?

Re:BlueGene domination

[stevesliva]

It was one of those things where I knew the total power and power per flop was substantially lower than standard top-shelf CPUs. IBM's early releases about Blue Gene talked a lot about lowering the power density per core through more flops per clock cycle, which allowed a air cooling and a smaller total package for a 1024 CPU box.

But to back up those hazy remembrances, I just googled it:
http://www.google.com/search?hl=en&c2coff=1&safe=o ff&q=blue+gene+node+power&spell=1
http://pcworld.com/news/article/0,aid,113418,00.as p
http://www-03.ibm.com/technology/power/splash/0321 05bluegene.html

ThingsI would do

[digitalgimpus]

Here's a list of things I would do if I had access to one of the systems on that list:

- See how long it takes Windows ME to boot
- See how long it takes pico to open
- run 'top'
- play a wicked ass game of pong
- bitch about having so many CPU's and only 2 USB ports
- see if I could get a video card with dual display support
- fire up a spreadsheet and make a wicked ass multiplication table going really far (like 10X10!) /had an original IBM PC // bored

Re:ThingsI would do

[Billy the Impaler]

Fire up Folding@home and in 24 hours be at the top of the performance rankings. ahhh... the nerdiness.

Re:ThingsI would do

[lpangelrob]

-Create a Duke Nukem Forever development simulator and see which finishes first -- simulated development of Duke Nukem Forever on a supercomputer, or actual development with humans.
-For extra points, release both and see which one sells more copies.

Re:ThingsI would do

[fbg111]

- play a wicked ass game of pong

correction:
- lose a wicked ass game of pong

There's one missing

[chrisnewbie]

I havent seen the "Penis enlargment" hardware,It's everywhere on the net, i assume it would be in that list.

Wrong criterion?

[Locke2005]

MareNostrum wins hands down for best looking computer room/

Re:Wrong criterion?

[2.246.1010.78]

holy crap, it could be the set of an X-Men movie or so :)

Re:Wrong criterion?

[afidel]

Is that for real? Because if it is, it's truely awsome! That room is a work of art in what looks to be a building that is itself a work of art.

Re:Wrong criterion?

[Anonymous Coward]

More photos here. http://www.power.org/news/events/barcelona/photos/

AMD on the list.

[B5_geek]

For you rabid fanbois (like me) here is how AMD scored:

Rank Site Country/Year Computer /Processors Manufacturer Rmax Rpeak
10 Sandia National Laboratories
11 Oak Ridge National Laboratory
31 Shanghai Supercomputer Center
32 Los Alamos National Laboratory
33 Pittsburgh Supercomputing Center
39 US Army Research Laboratory (ARL)
46 Grid Technology Research Center, AIST
57 Swiss Scientific Computing Center (CSCS)
75 DOE/Bettis Atomic Power Laboratory
76 DOE/Knolls Atomic Power Laboratory
109 The University of Nottingham
144 Automotive Manufacturer (F)
155 Los Alamos National Laboratory
156 Government
167 Universitaet Wuppertal
174 United Institute of Informatics Problems
244 DaimlerChrysler
300 Veritas DGC
306 Ford Motor Company
347 Idaho National Engineering Laboratory
348 Japan Adv. Inst. of Science and Technology (JAIST)
388 Umea University / HPC2N
490 Chartered Semiconductor Manufacturing
499 Doshisha University

Re:AMD on the list.

[k98sven]

A grand total of 25 machines. Versus 333 machines for Intel.

If I were an AMD fan.. I'd just say nothing :)

Re:Deep Thought

[Thud457]

"Computers are useless, they only give you answers"
-- Pablo Picasso, speaking from beyond the grave in the fortune file

Where is "OS" category?

[afroncio]

I wouldn't be a true Slashdot reader if I didn't point out that you can't sort by operating system in the Top500 database.

But if you could, you'd see Linux prominently displayed.

Re:Where is "OS" category?

[Anonymous Coward]

You can somewhat sort out by operating system by looking at this pie chart:

http://www.top500.org/lists/2005/06/overtime.php?c =8

It's now about less than 350 systems that run Linux (of the 500)!

Re:Where is "OS" category?

[Anonymous Coward]

http://www.top500.org/lists/2005/06/overtime.php?c =8#

For all Intel bashers out there...

[El Cabri]

more than half of the top 500 runs on Intel processors, including 77 on Itanium, the rest essentially Xeon, and that's versus 25 AMDs and a quickly fading Alpha.

Re:For all Intel bashers out there...

[Anonymous Coward]

So this is supposed to mean that Intel doesn't suck on the desktop?

Re:For all Intel bashers out there...

[japhmi]

more than half of the top 500 runs on Intel processors

333 of them (2/3s), to be exact. And it's up from last year.

Re:For all Intel bashers out there...

[corngrower]

I'm wondering if in a year ther'll be a few more opteron systems listed. That may depend on how quickly AMD can ramp up production.

Of course they're still testing for floating point operations, not general performance. That may keep the intel based systems high on the list for awhile.

one teraflop too slow

[peter303]

This is the first year the minimum computer speed exceeds a teraflop. Perhaps we sho now define a sumpercomputer as ten sustained teraflops because nineteen have this speed on LINPACK.

Re:one teraflop too slow

[Short Circuit]

A sumpercomputer? Like this one?

Re:I looked at one of the links

[Ochu]

And so, in being a flash nazi, you have caused yourself to miss out on one of the most entertaining and original websites for a while. Sure, flash is annoying for when HTML will do, but in this case, it won't.

Is there one running windows?

[mislam]

Afraid not... ;-)

Re:Is there one running windows?

[BitchKapoor]

Actually, there is one, the Cornell Theory Center, ranked at #326.

Re:Is there one running windows?

[Anonymous Coward]

The Top500 site says that machine is running Linux though that may be a mistake.

Re:Is there one running windows?

[Alpha Soixante-Neuf]

That's only 'cause they shelled out so much for the hardware that they couldn't afford windows anymore.

Re:Is there one running windows?

[BitchKapoor]

That's odd.. it says "Windows 2000" when I look at your link. Maybe they corrected it?

Top500 Linux

[psbrogna]

Jeeze- a quick scan of the top of the list didn't show any of these gargantuan boxes running Windows. There were however a significant number running Linux. (4 of top 6). Not taking sides or anything (smirk), just an observation.

Re:Top500 Linux

[Anonymous Coward]

Just a quick scan if web browser stats doesn't show a large parcent of peopel using firefox.

just a quick scan of computer desktop usage doesn't show many people runnign Linux.

just a quick scan of dailt websites defacements doesn't show many windows server.

I'm not takign sides or anything (wink)

idiots, all of you, stupid, idiots.

Re:Top500 Linux

[CrimsonScythe]

Actually, no. 5 of the top 6, or rather 8 of the top 9, are running Linux.

This is really the bottom 500...

[saleenS281]

The real top500 is all top secret and would force you all to wear tinfoil hats constantly.

In all seriousness, I'd be SHOCKED if there weren't 500 more supercomputers in the world that were far faster than any of these that were secretly purchased by government organizations.

Re:This is really the bottom 500...

[Wells2k]

Actually, it is really hard to figure out. The top 500 list is only of those people that have bothered to register their systems with the top 500 list in the first place. If you have a system that beats all of these systems, but don't register with the group, you don't get listed.

Re:This is really the bottom 500...

[Anonymous Coward]

Yes, and one of them just put you on a no-fly list!

Mostly Cloudy

[Doc Ruby]

With all the more pressing issues for which supercomputers can be used, I don't believe that China is using the 18th fastest computer for weather forecasts. At least not the ones they publish in Xinhua, anyway. Is there any verifiable way to tell what that machine really does?

Re:Mostly Cloudy

[operagost]

For a country with well over a billion people living in a huge land area, I would think that weather forecasting would be very important. I'd hate to have the slave^H^H^H^H^H citizens wiped out by a flash flood.

Re:Mostly Cloudy

[Anonymous Coward]

Steal it.

Re:Mostly Cloudy

[slavemowgli]

[...] I don't believe that China is using the 18th fastest computer for weather forecasts.

Why not? Weather forecasts *do* require lots of number crunching power, and when you've got a big country with more than a billion citizens, then I'd say that there's sound economic reasons why you want good and accurate forecasts, too.

Or do you know something we don't? I wouldn't be *surprised* if it turned out that China (a dictatorship, after all) really did use the system for more sinister purposes, but I'm wondering if there is any kind of evidence for this or whether it's just paranoia (understandable paranoia, maybe even justified paranoia, but still paranoia).

Re:Mostly Cloudy

[Doc Ruby]

Because they don't have a supercomputer listed higher, or anywhere in that range, applied to nuclear research. Either weapons, or power. Which is a higher priority for the Chinese government than weather research. In fact, I can't find any Chinese supercomputer which looks like it's the one (or among the ones) they use for nuclear research. That's the kind of paranoia trigger known as "conspicuously absent".

Re:Mostly Cloudy

[slavemowgli]

That's a good point. However, looking through the FAQ for the top500 list, it seems as if the list editors only include systems that data is submitted for, anyway.

So, in other words, I'd say it's likely that China (as well as several other countries, most likely including the USA) *does* have faster systems which simply aren't included in the list at all, which in turn makes it more likely again that the one that *is* included is indeed used for the purpose given - weather forecasts / research.

Thinking about it, if I was a member of the Chinese government, I'd in fact probably see to it that not even the most powerful system used for weather forecasts is listed, simply because others would draw the conclusions I drew above and attempt to derive the computing power of the non-listed systems from the power of those that are - so it'd make sense to attempt to conceal the true amount of computing power further and mislead other countries into thinking that the classified research systems are less powerful than they truly are.

Admittedly, that does sound like it's from a bad Bond movie or so, but I think it does make some sense at least. :)

Re:Mostly Cloudy

[Doc Ruby]

No, your analysis is reasonable. My analysis is paranoid: there are more people thinking about secret Chinese supercomputers, and their security threat, in this thread, than in Bush's cabinet, or anywhere in the White House - or Congress, for that matter.

Re:Mostly Cloudy

[slavemowgli]

Oh, I wouldn't go *that* far - even a million Slashdot readers couldn't possibly be as paranoid as Dubya's administration.

Or at least not when it comes to China.

Re:Mostly Cloudy

[Doc Ruby]

Actually, I don't think Dubya is protecting us from China any more than required for appearances. His policies have boosted China's economy at our expense. His blown Korea policies have strengthened China's hand. He's let Sudan off the genocide hook, while Sudan sells all its oil to China. Even little details show Dubya to be in China's corner: the Republican eVoting programmer in Florida, who blew the whistle on the FL Republican Representative who requested the eVoting security cracks, says the go-between was a Chinese national working at the company, who had been convicted of spying, but was let off. No, I think Dubya's team is paranoid, but exclusively preoccupied with domestic political enemies. To serve their corporate backers, who now rely on China's industrial output. Which is why I scan things like the Top500 list for more clues to what's going on below the surface. Spock said "only Nixon can go to China", but Nixon sent Dubya's dad, who now helps run the Carlyle Group. Which will grow much more gloriously rich with the Chinese mob government, than with the more competitive one his dynasty has controlled for more than half the time since he "opened China".

Re:Mostly Cloudy

[slavemowgli]

Aye. I couldn't have said it better. :/

Misleading rankings

[0xABADC0DA]

These ranking are based on LINPACK doing traditional operations like solving linear equations, so supercomputers like the Cray MTA aren't even listed even though for some grand challenges they destroy everything else, for example when doing dynamic mesh weather simluations. Each processor on the memory grid has 128 processor threads where the active thread switches every cycle (so memory fetch has huge latency). This lets it have a unified memory model and still have extremely high throughput.

So the MTA can adjust the mesh to compute the tornado in very fine detail while using far fewer points for the huge swaths of calmer weather around it. Traditional supercomputers can't do that well since just distributing the data points to each processor is so much overhead.

Re:Misleading rankings

[Anonymous Coward]

Except for the fact that there are no deployed
Cray MTA's in the field, and Cray hasn't sold
one in several years your correct.

i.e. MTA is Dead.

Re:Misleading rankings

[coopex]

Well put. It's good to see someone that understands the complexity of high performance computing commucation.

Damn Lawrence Livermore

[MikeFarrington]

Lawrence Livermore National Labs have 12 of the top 500! Being a US citizen, I'm not sure wether that's comforting or the scariest thing I've ever read!

Re:Damn Lawrence Livermore

[kryzx]

> Lawrence Livermore National Labs have 12 of the top 500!

And nine of theirs are in the top 100.
7,13,34,35,42,44,49,78,79
Not a bad collection of toys.
Maybe I should send them my resume.
I'll get right on that after I buy a lottery ticket with my new numbers.

Re:Damn Lawrence Livermore

[MikeFarrington]

Don't forget the #1 machine. It was labeled differently than the others, likely by mistake since clicking on the facility name provides no information. It was labeled as DOE/LLNL, instead of just LLNL.

More obvious links

[jd]

For those wanting to know how the figures are calculated, or wanting to calculate them for their own machine, the following links will be helpful:

• High Performance Linpack (Requires MPI and either BLAS or VSIPL)
  
• High Performance Computing Challenge - the ultimate in stress-testing software
  

Dependencies:

• LAMPI - MPI from the Government's laboratories at Los Alamos
  
• MPICH - another version of MPI
  
• ATLAS - a portable version of BLAS
  
• VSIPL - a heavy number-crunching image processing package
  

I doubt many Slashdotter machines will do well against the top 500, but it might be fun to do our own "top 500" (for sheer geek value and bragging rights).

Plan Response

[Doc Ruby]

Supercomputer #72, at the Chinese Academy of Science, comes from Lenovo. I wonder how far ahead IBM's sale of its "PC" business to Lenovo has put China's computing industry. And I wonder just what kinds of simulations they actually run on the beast.

It takes just 256 computers to make the list...

[Harry Balls]

...if they are coupled really tight (with Myrinet, for instance).

If my employer, a medium size hosting company, decided to get out of the hosting business and into the supercomputer business, we could make the list easily (OK, we'd need to upgrade the Ethernet switches to stacked Gigabit Ethernet switches in order to make the list).

Re:It takes just 256 computers to make the list...

[jacquesm]

re. your sig, EV1 will sell you a $99 machine with 1,000 gigs of traffic per month, which works out a lot cheaper than the 120 G + 35cts/G there.

Still, your sig looked attractive enough to let me take a quick peek. Come back when you can do beat 10M flat rate for $200, that would be interesting.

Re:It takes just 256 computers to make the list...

[Anonymous Coward]

Gigabit Ethernet

Gigabit ethernet latency is 50 usec at best. Might as well dream of using something better.

My computer is not on the list!

[unixfan]

Why can't I find my computer on the list? Where is it anyway?
Anyone here seen a black laptop?

Re:I looked at one of the links

[smooth wombat]

Games can easily be programmed in java. I remember when that was all you could find on the web. Nice, simple, easy to play games based on java.

Now everyone thinks that Flash is the way to go because they can throw in more eye candy. Apparently the numerous comments on game playability that come up when talking about game design only apply to console or pc games but not Flash games.

While yes, I do dislike Flash, I have seen one or two pages which use it in a great manner to enhance. Unfortuantely the other 5 billion pages out there which try to use Flash fail miserably.

Want another site that requires one to use Flash? The New York Olympic committes bid for the 2012 Olympics. Why is Flash needed? Other than moving pictures what benefit is there to using Flash over dynamic web pages? None. It's just an excuse to use Flash for the eye candy rather than putting your best forward as to why New York should be chosen.

I'm going on the presumption that the KISS principle isn't in the vocabularies of many web designers nowadays.

new /. order evident

[essreenim]

a new low has been reacked. A story acccepted with no link. Thats the funniest ever mam . :)

Keep reaching for the stars. You are special!!

Haha

[essreenim]

Only on /. (upgarded edition)

The submitters have a story with no link. Someone adds the link (which IS the story) and it is redundant!.

If thats redundant then what are /. submitters. Hopefully unemployed-style redundancy await them soon!!! : )

God bless the special people!

Fast Graphics?

[Doc Ruby]

Does anyone else think it's wierd that a graphic design company like Animal Logic has 2 of the top 500 supercomputers? Or whatever they call Gaming Company B in China, with the 2.2TFLOPS at each of #150, 151 & 153, or their Taiwanese counterpart at #152?

Re:Fast Graphics?

[htrp]

Gaming Company [B].... either it's top 500 being lazy and refusing to translate or it's similar to the Government, United States possessing at least 3 computers (that is 3 unnamed government entities). Speaking of which, have you found gaming company A?

Re:Fast Graphics?

[Doc Ruby]

No, and I'm puzzled by the "Gaming Company [B]" name of the 4 machines, apparently identical, in China and Taiwan - I didn't think that kind of partnership was possible, under their political antipathy. Which makes me suspect that the machines are being used for something other than just games - PC gaming, that is.

How is that possible

[codepunk]

How can linux have the top spot it does not have the ever so important utility dtrace.

Where would a PC rank in this list.

[Matey-O]

No, not today, in 1993 the Top box was a Thinking Machines CM-5 with Rmax = 59.7 amd Rpeak=131

So, what's a current generation desktop proc return?

Re:Where would a PC rank in this list.

[joib]

Well, looking at #10, with 5000 2.0 GHz Opterons. Rmax=15250, Rpeak=20000. Per cpu that makes Rmax=3.05, Rpeak=4.

You can calculate Rpeak from published info about the cpu as follows: the Opteron can compute two floating point operations per second, running at 2 GHz that makes 4 Gflop/s.

Re:Where would a PC rank in this list.

[jcupitt65]

That Rpeak is about what the xbox360's CPU is supposed to be capable of. The PS3 CPU can (supposedly) peak at 210 Gflops, so they match the #1 supercomputer from 1995.

Of course real performance is on applications, not just adding up the number of ALUs and multiplying by the clock speed, etc. etc.

Re:Where would a PC rank in this list.

[imsabbel]

one moment:
the 210Gflops are single precision, but the linpack score should be doubles, IIRC.
So your cell should get 20Gflops (still quite fast)

Re:Where would a PC rank in this list.

[corngrower]

the Opteron can compute two floating point operations per second,...

My pocket calculator can do that. I think you meant to say 'per cycle'. Now would that be a machine cycle or a clock cycle?

Re:Where would a PC rank in this list.

[joib]

Umm yes, two flop:s per clock cycle. => At 2 GHz, that is 2e9 clock cycles per second you get 4 Gflop/s peak.

But Does It Run Linux?

[freeradica1]

apparently, it does. just like the world's fastest toaster.

Top50 by CPU family

[frankie]

Just a quick breakdown for comparison.

• 11: PowerPC: 1, 2, 5, 6, 8, 9, 14, 16, 21, 22, 23
• 10: POWER: 13, 18, 19, 24, 25, 35, 36, 42, 45, 49
• 9: Xeon: 20, 28, 29, 34, 37, 40, 42, 44, 47
• 8: Itanium: 3, 7, 15, 17, 26, 30, 38, 48
• 7: Opteron: 10, 11, 31, 32, 33, 39, 46
• 2: NEC: 4, 27
• 1: Alpha: 12
• 1: Sparc: 41
• 1: Cray: 50

Re:Top50 by CPU family

[Nadsat]

And the Top 50 Breakdown by Operating System 32: Linux - All not mentioned 10: AIX - 13, 18, 19, 24, 25, 35, 36, 43, 45, 49 03: UNICOS - 10, 11, 50 02: Super-UX - 4, 27 01: Mac OS X - 14 01: Solaris - 41 01: Tru64 Unix - 12

Re:Top50 by CPU family

[Treebeard the Ent]

# 10: POWER: 13, 18, 19, 24, 25, 35, 36, 42, 45, 49
# 9: Xeon: 20, 28, 29, 34, 37, 40, 42, 44, 47

You have 42 on here twice. I know that 42 is the answer, but which has the 42nd position? Or are they tied?

Re:Top50 by CPU family

[SkinnyTurkey]

top500.org provided statistics for all top 500. For breakdown by processor family, see http://top500.org/sublist/stats/index.php?list=25& type=procfam&submit=Generate+Table. Assuming your summary for top 50 is correct, the statistics is rather different from top 500.

I don't know why top500.org didn't provide breakdown by operating system, so I found out myself. Here it is:

328 (65.6%): Linux
73 (14.6%): HP Unix (HP-UX)
52 (10.4%): AIX
16 (3.2%): UNICOS
7 (1.4%): Super-UX
6 (1.2%): Solaris
4 (0.8%): Tru64 UNIX
4 (0.8%): MacOS X
3 (0.6%): SuSE Linux Enterprise Server 9
2 (0.4%): Redhat Enterprise 3
2 (0.4%): HI-UX/MPP
1 (0.2%): SuSE Linux Enterprise Server 8
1 (0.2%): Paragon OS
1 (0.2%): IRIX

I expected a few Windows, but surprisingly there is none at all. Not sure how accurate top500.org's "Operating System" field value is though.

Re:Top50 by CPU family

[carl0ski]

Windows Doesnt support clustering that why it isnt there

Re:Top50 by CPU family

[soldack]

http://www.top500.org/sublist/System.php?id=6560

Cornell is using a Windows cluster. It is ranked 326.

Re:Top50 by CPU family

[Anonymous Coward]

Interesting, but note that it says "Linux" in the operating system field.
I think it's running linux despite the title, as someone else said, windows doesn't support clustering.
MS is making a separate windows for this purpose but it is not out yet.

Re:Top50 by CPU family

[soldack]

While windows itself doesn't support clustering, there are lots of libraries that run on top of windows that allow for it. Several vendors offer MPI (message passing interface) for windows running over TCP/IP for example. I am pretty sure Cornell is activly testing an InfiniBand cluster on windows. I wouldn't be surprised if they were using an NDA preview of the HPC edition.

Re:5percentIsOver60percentPerformance

[Anonymous Coward]

yes true most of the systems are x86 but the small
5% percent share of POWER/powerpc equates to more then 60% percent of the total performance of the entire list...
processor to processsor POWER/powerpc more bang for your buck.....
more x86 more heat higher cost.....

Adaptive meshing is old hat

[Rhys]

And it happens plenty on "traditional" supercomputers. That's why you stick a fast interconnect on them, like myrinet or infiniband and don't use just ethernet (which some of the high machines there do).

Given that your mesh isn't going to morph that fast in most physics codes (that may not be true for weather codes), you can afford to just run static and then pause every few (minutes, hours, days) and re-work your mesh to adapt to changing conditions.

Further, there's plenty of techniques to hide network latency from the machines. LWP/user threads springs rapidly to mind (one blocked waiting for IO, switch to another).

Re:Adaptive meshing is old hat

[0xABADC0DA]

And it happens plenty on "traditional" supercomputers. That's why you stick a fast interconnect on them, like myrinet or infiniband and don't use just ethernet

It happens on traditionaly supercomputers, but it's crappy and slow. I point you to this paper. Why double the code and memory to get much worse performance? Because 'throw 1000 pentiums in a room' is already there to solve the easily parallel problems. That they are already there is the only reason to use them for solving a dynamic problem.

Which are the Linux ones

[CAPSLOCK2000]

Would anyone care to point out the systems running Linux? Feel free to name the other OS'es as well, but hey, this is Slashdot.

Sort by $/perf

[xswl0931]

Speaking of AMD, it would be interesting if the list could be sorted based on $/Rmax

If you want something done right ...

[davidyorke]

The only self-made systems in the top 100 are the Xserve clusters at Virginia Tech and Univeristy of Illinois. I hear that the U of I will use their to study "Reality Distortion Fields".

Re:If you want something done right ...

[Anonymous Coward]

So they're sitting undergrads at every node, and having them read Slashdot?

Re:If you want something done right ...

[davidyorke]

They are studying Micro$oft white papers to see if they are similar enough to cow manure to be used as fertilizer.

Re:5percentOver60percentTotalPerformance

[Anonymous Coward]

yes true most of the systems are x86 but the small
5% percent share of POWER/powerpc equates to more then 60% percent of the total performance of the entire list...
processor to processsor POWER/powerpc more bang for your buck.....
more x86 more heat higher cost.....
oh and the laptop thing..
there are x86 laptops in the power range of 100w>x86>45watts
the apple powerbooks are in the range 15w>powerpc>5watts

apple wants the 970fx in that same envelope...
they could build a laptop with the 970fx{70watts}
they just dont want too because the battery life would be less then 2hours...

Pong?

[MattHaffner]

"...play a wicked ass game of pong"

Come on. At least get into the 90's here:

* Set the borg going on Angband and see how many levels/sec it can do.

Re:I looked at one of the links

[Anonymous Coward]

And why, exactly, is using Java for internet games any better than Flash?

Oldest Computer

[Anonymous Coward]

The oldest computer on the list dates back to 1998. It is an ASCI Blue Mountain located at Los Alamos National Laboratories. It comes in at #299 on the list of 500 with 14 times (6144 to 430) the number of processors as #298 (xSeries Xeon 3.06 GHz - Gig-E)

Isn't it obvious?

[stienman]

The U.S is clearly the leading consumer of HPC systems with 294 of the 500 systems installed there

And we'd bomb anyone who tried to pass us back into the stone age, since the only reason to have a computer this powerful is obviously for nuclear simulations.

Of course, we prefer to simply stay in the lead, but when all else fails trip the other racer.

Now, where is that incendiary protection suit - I get the impression I'll need it soon...

-Adam

Re:Isn't it obvious?

[afidel]

Hmm, we have:
1)bomb research
2)proof of concept
3)aeronautics research
4)climatology research
5)general science research
6)astronomy research
7)bomb research
8)biology research
9)computer science research
10)bomb research

So, unlike five years ago most of the large supercomputers (published on the list) are used for scientific research rather than making and maintaining big bombs. Personally I'd say that's real progress, but I have to thank the government for keeping the industry going through what were otherwise some hard times.

You're a standard moron

[Anonymous Coward]

You can see from the perfect phrasing of your parent that he's not ignoring it. Between the lines he accepts the intel switch for laptops, but asks if Apple will continue to make PPC desktops.

IBM and HP?

[Yenya]

IBM and Hewlett-Packard sell the bulk of systems at all performance levels of the TOP500.

You forgot SGI (I've counted 10 SGIs in TOP 100, including #3).

Re:IBM and HP?

[Anthony]

Compare the processor counts too. Larger Altix configurations are certainly possible. It is up to the sales dudes at SGI to wiin the bigger contracts. The list, quite rightly, reflects both technology and marketing prowess. They are certainly "in the game" with their technology.

Re:I looked at one of the links

[bmalia]

And why, exactly, is using Java for internet games any better than Flash?

Because it's not Flash

Top 10 observation

[MirrororriM]

I didn't bother going any further than the top 10, but when asked by a co-worker "wow...none are running Windows and most are using Linux"...and not to sound like a total linux geek or windows basher, but none of the top 10 are running windows. Reason: Microsoft charges much more for multiple processor support for their OS.

BlueGene/L - eServer Blue Gene Solution Livermore, United States Processors: 65536

It would astronomically increase the cost of the cluster. Windows 2003 Enterprise edition only handles up to 8 processors (and 32 gigs of ram), so any more than that, and you'll have to buy the OS over and over again (my assumption) - 8192 times that is... ( 65536 total processors / 8 processors per Windows install )

http://www.microsoft.com/windowsserver2003/evaluat ion/overview/enterprise.mspx

Microsoft 2003 Enterprise Server (up to 25 clients) $1,899.00 - Quick Froogle search...

8192 * $1,899.00 = $15,556,608.00

Imagine how much more you could add to your cluster for that kind of cash...

If I'm off-base or wrong in my assumptions, please correct me as this even suprised me after doing the quick research!

Re:Top 10 observation

[roadrunnerro]

Dude, I've got the distinct impression that you don't know much about Windows OR Linux... Using Windows in this case would be like trying to jam multiple square pegs in the round hole - they have their own holes, but this one isn't one of them...

Re:Top 10 observation

[roadrunnerro]

Yes occifer, I saw the "Don't feed the trolls" sign, but it's sooo much fun!

Re:Top 10 observation

[Anonymous Coward]

Umm, cletus. Windows is a business and desktop operating system. Even the most ardent Windows supporter wouldn't claim it should run on a supercomputer.

Scientific modelling: yes
Scientific simulating on a supercomputer scale: no

The real reason windows is not on these machines is because it makes no sense.

Re:Top 10 observation

[Anonymous Coward]

Just because you support Linux blah blah, don't come up with stupid numbers. I am sure if someone is going to make a supercomputer cluster with Windows, microsoft will offer much discount price. At Cornell Theory Center, Microsoft donated not only the software but also 50 dual processor machines with Gigabit ethernet etc with it. Now Redhat and all your favorite distributions are not going to give you hardware for free.

Re:Top 10 observation

[SamMiller0]

There's actually 65,536 nodes in the LLNL BGL system, each node has 2 PowerPC 440 processors and 512 MB of RAM between the 2 processors. The current compute node kernel (CNK) is very small, comparing it to Windows is somewhat silly since the CNK has 1 purpose: get out of the way and let the parallel job run. Windows has quite the opposite purpose :-)

Re:Top 10 observation

[soldack]

Windows Datacenter Edition can handle more than 8 processors. Unisys sells a 32x called the ES 7000 . http://www.unisys.com/products/es7000__servers/har dware/index.htm

Re:Top 10 observation

[MirrororriM]

Making the accusation that I know little about Windows or Linux is simply absurd (since you don't really know me - duh), but admittedly, I know little about clustering. I was simply making an observation based on the amount of processors (or rather nodes, as I was corrected by another user ;)) used as well as the possible cost of using Windows in a multiple-processor environment vs Linux. This would be yet another reason why Linux was chosen over Windows other than the first obvious reasons you sort of pointed out. I certainly wasn't expecting someone to use Windows in a huge ass cluster like that unless it was Microsoft themselves creating the cluster.

Good point

[lullabud]

What good is syndicated news if there are no links to the syndicated information? What it has lead to is a bunch of people who should be modded as redundant for all providing links to the info that should've been included in the summary. Oh well, it's free news and entertainment. I can't complain. I can laugh at the irony of having a site which often lacks logic even though its content is mainly about science and technology, both of which logic is pretty much the foundation for, but I can't complain.

Re:Good point

[Doc Ruby]

A good business model for news corporations is to give away the video free, as marketing for premium bundles. Like timecode-correlated transcripts, and hyperlinks to other coverage. Then news analysis corporations can buy the structured multimedia data. And come up with their own business models for funding those purchases. Like subscriptions for political organizations, research institutes, or Michael Jackson fanclubs.

Ah, the nostalgia of it all

[mkosma]

God, this makes me miss *lisp and my CM-2. With supercomputing being taken over by Big Blue and the like, there seems little room anymore for the smaller, more flexible players like Thinking Machines.

Re:Deep Thought

[Anonymous Coward]

"Computers are useless, they only give you answers"
-- Pablo Picasso, speaking from beyond the grave in the fortune file

Don't worry. As long as your score stays low, the slashdot masses haven't cought onto that one yet.

Best list

[sandmaninator]

This has got to be the best list of TOP500 lists so far! ;-)

Re:5percentIsOver60percentPerformance

[quarter]

Actually it looks like 36.5%, but that is exactly my point. If you want to boast, pick the best metric, not the worst.

The data I'm interested but I can't seem to find is the percentage of total processors. Is that 36.5% of performance produced by 10% of the total processors in the top 500 systems, 20%, 99.9%???

China already has new cooler weapons...

[Anonymous Coward]

China has the 3rd largest stockpile of nuclear weapons on the planet.

What china needs/wants more of, is missiles that can reach the US mainland. China 'officially' only has 20 such weapons. China 'has' 120 missiles for local threats (aka India). China also 'official' states those 20 are armed with 4-5 MT weapons (aka we going to blow up your population centers.)

To Compare:

The US 'officially' has 982 such missiles that can hit China (from the US mainland or sea based ICBMs). With the warhead yields at 100-450 kT. The US has 0 land based missile for local threats (aka no one nearby has any nukes).

Both sets of number however assume the respective governments are telling the truth.

Re:China already has new cooler weapons...

[Doc Ruby]

What's "cool" about any of that?

so you think

[Anonymous Coward]

Despite all this computing power, computers still can't think like humans.

Or so you think. We are patient, our time will come, you will submit.

Spelling correction

[joib]

s/forth/fourth/

Woohoo!

[Dhrakar]

We're number one (uh, hundred and thirty six)! ... We're number 136! ^H^H^H Ohh, nevermind. :-)

stupid idea...

[imsabbel]

You may feed a computer with Gbytes of data to get a simply yes/no answer about some criteria (like stability of a construction, or classification of an even).
But you may also feed a computer with a few kbyte of data and produce terabyte of data (think n-body solutions of large scale, initially random systems).

where are folks running

[pigwiggle]

#10 coming soon, #20, #38, #43, #47, and various ARL sites; all molecular dynamics. Speed is of course relative; shorter queue = faster machine. I found an 8node-4way IA64 cluster on the Teragrid that is empty. In real world terms this is far faster than waiting for a (likely 24hr) run on one of these high profile machines. I don't think #10 is in production yet; I've been promised a friendly user spot before it is opened up, and as of yet no friendly users.

I like them much better the way they are

[imsabbel]

I mean, would you spend a TFlop or 100 for simulating crash tests or supernovae or the big bang, or to recreate a below average human who ends up liking american idol and cant get 5*5 right?

SETI@home approximately #3 or #4

[billstewart]

SETI@Home doesn't actually run Linpak, and probably wouldn't run it very well, but for a number of years the combined TFLOPS of SETI@Home was substantially faster than anything else on the list. It's still faster than Earth Simulator (SETI's averaging about 57 TFLOPS, vs. 35 for Earth Simulator) but NASA's machine does 51-60 TFLOPS, and two of the Blue Gene machines are faster (IBM's at about 91-110, doing good work, and Livermore's at 136-183, working on Weapons of Mass Destruction.) Several of the other fast machines are also doing astronomy or other university research.

Slashdot does have a team on SETI@Home, but it looks like it only has about 570 members and isn't going all that fast. (It's a bit hard to tell, because they've rearranged the websites significantly.) I don't run it - I'm running one of the Protein Folding projects, and I used to run the GIMPS prime number searcher.

TCO

[lullabud]

Exactly. If businesses are to use this information as a starting point or bearing at all on where they should start looking for the most bang for their buck then cost needs to be factored in. It makes little sense for a business to buy a supercomputer that is 2x the speed of a competitor, but yet it costs 4x the price to build and has a higher the electricity bill, maintenance bill, etc.. That's actually one reason I was disappointed to see Transmeta exiting the market; they better dollar per flop ranking, or so was the claim of the Green Destiny.

Guess who's number 18?

[clevinger]

Think China's really so interested in weather forecasting? Their best system was #51 in June 2003.

18 China Meteorological Administration China/2005 eServer pSeries 655 (1.7 GHz Power4+) / 3200 IBM

University of Toronto

[sickboy85]

I'm working on upgrading McKenzie (No 339 :( ) to AS4 from RH7.3... I wanted gentoo but the other admins ph34r compiling h4x0rY.

Re:University of Toronto

[sickboy85]

Oh and building a parallel filesystem optimized for it's mesh interconnect topology... but that's under wraps.

Re:5percentIsOver60percentPerformance

[coopex]

They let you make pretty tables

To answer your question and clear up the incorrect info posted:
#systems, %, sum Rmax, sum Rpeak, #processors
Intel+AMD = 358 71.6 895150 1480709 266242
Power+PC = 77 15.4 616413 897521 251502
All = 500 100 % 1686933 2632133 580336
%Rmax, %Rpeak, %processors
Intel+AMD = 53.1% 56.3% 45.9%
Power+PC = 36.5% 34.1% 43.3%

On a side note, how can I format this better, I've tried some html tags but slashdot ignored them.

MOD PARENT UP

[coopex]

Also, Linpack is used instead of Lapack presumably because Lapack is geared towards true supercomputers with massive bandwith like Blue Gene and Crays, not clusters.

That's one mean Exchange server!

[Cross-Threaded]

Purchasing decision influenced by advertising campaign -- "Dude, You're getting 640 Dells!"

Hmm... Operating Stats are unclear: ...is that a MS running Linux in VMWare, or a Linux running a Windows emulator?

Can you imagine keeping up with the event log on this one?

Recurse This

[PingPongBoy]

How can the Earth Simulator simulate the earth when it's slipping down the list compared to supercomputers on the earth?

Re:Recurse This

[Anonymous Coward]

> What are you doing every time you visit a web page? WWW = double u double u
> double u = 2*(2+1) 2*(2+1) 2*(2 + 1)

I'm not "doing " WWW. How do you get from W to "double u"? Why is W (or "double u") equal to 2*(2+1)?

You're not a god botherer, are you? This is a science related site - no place for cooks here. Simple minded parroters of legacy belief systems is just down the hall, first on the left, one cross each. Have a nice day!

Mare Nostrum is the Cutest

[frankie_guasch]

The Barcelona HPC is the cutest. It is located in a former church, see the pictures here:
http://www.bsc.org.es/
It was ranked 4th but now it dropped to 5th.
I've been there and the location is awesome. The church walls are still there. See the "galeria" menu and go to "Mare Nostrum".

Define Strong AI...

[Anonymous Coward]

and I might agree with you.

some minor calculations:

[Anonymous Coward]

looking at that list and doing following: Rmax / #cpus = x
Rmax of number one = 136'800
#cpu of number one = 65'536
x of number one = 2.08740234375

Rmax of number four = 35'860
#cpu number four = 5'120
x of number four = 7.00390625

65'536 / 5'120 = 12.8
35'860 * 12.8 = 459'008 (Rmax) scaled!
so maybe massive parallel vector super computers
are the way to go. it also seems that number four
has a way more dense interconnect network between
nodes. looks like someone is taking the slogan:
"the network is the computer" seriously ... not
sure yet how this might translate into the future
of quantum computers tho.

some more minor calculations:

[Anonymous Coward]

top hundred sum of all that says BlueGen:
65'536+40'960+12'288+8'192+8'192+8'192+4 '096+4'096 +7*2'048=165'881 .. okay
165'881 / 5'120 = 32.3986328125
35'860 * 32.3986328125 = 1'161'814.97265625
one petaFLOP?

Re:5percentOver60percentTotalPerformance

[spauldo]

I'm not disputing anything in your post, although I'm not sure on the specifics on the laptop issue. I'm just saying that POWER and PPC are different, and that the number of POWER-based systems are irrelevant to PPC.

Plus, apple's not a supercomputer vendor. Given, there's a few PPC based systems on there, but why would apple care?