Slashdot Mirror
Japan Wants to Build 10 Petaflop Supercomputer
deepexplorer writes "Japan wants to gain the fastest supercomputer spot back. Japan wants to develop a supercomputer that can operate at 10 petaflops, or 10 quadrillion calculations per second, which is 73 times faster than the
Blue Gene. Current fastest supercomputer is the partially finished Blue Gene is capable of 136.8 teraflops and the target when finished is 360 teraflops."
Well I want a Stargate, but that doesn't mean I'm gonna get one. I bet OpenOffice.org will still take 5 minutes to start on it.
The supercomputer will be pocket-sized and ran on two AA batteries.
I see they are upgrading to get ready for Longhorn.
These 136.8 teaflops could have been avoided if the proper specifications were used before hardware development and programming began. Essential tea technical info.
I actually had to think for a minute "what is a teaflop?" then I realized it was a misspelling.
BlueGene/L is the fastest super computer at the moment; however, BlueGene/C (which, for the record, I'm working on as part of my PhD) will be finished very soon (it was supposed to be out of the foundry by the end of August, but the project is running slightly behind schedule). I'm told there are, as yet, no plans to publish any performance benchmarks.
To make laws that man cannot, and will not obey, serves to bring all law into contempt.
--E.C. Stanton
If funding runs out for this one, they'll end up with a 1 Belly-flop supercomputer
*ba-dum-dum ching!*
READY.
PRINT ""+-0
...and I want a pony.
Guess which two things aren't happening anytime soon?
What the hell is "Wot"?
I mean seriously... Doom 4 isn't even out yet.
**insert favorite profound quotation here**
No sure, Jo... But I think a petaflop is a pedicure gone bad. :)
Big deal, the white mice have had this beat for years...
Japan wants to develop
Japan wants a lot of things now doesn't it. Well, Japan will just have to be a good little country and maybe Santa will come.
The top news story of the hour:
Microsoft chairman, Bill Gates, has announced yet a new version of the Windows Operating system. Trying to take advantage of the obvious new market of supercomputers, the computer giant is ready to release Windows SC. The new operating system, designed to beat the Japanese domination in computing power, as well as the Russians in spam-distribution, will link all computers running the operating system into one giant spam^H^H^H^Hcommercial marketing distribution center.
Luke
----
Tired of answering tons of basic computer questions for friends and family? Send them to ChristianNerds.com instead!
There is the next step up: exaflop..
Seven, but the last one doesn't count.
japan is thinking "but we just bought this computer, it's obsolete already? shit a brick!" anyone in the market for a slightly used supercomputer?
Don't forget Beowulf clusters!
I can't believe there will be a computer that will actually play Diakatana at 1200x1600 in my lifetime
Optimized Gentoo without the need for a geek strung out on Mountain Dew.
Every mans' island needs an ocean; choose your ocean carefully.
The Japanese are really sensitive about the whole "small penis" thing.
--They say only a fool looks at the finger pointing to the sky...
What the hell is "Wot"?
:)
Read your comment above out loud and I'm sure you'll figure it out
"Who says nothing is impossible? Some people do it every day!" - Alfred E. Neuman
yeah... um... so I'm guessing OpenOffice would at least startup semi-fast on that machine
I can just picture the case mod.
Operator, give me the number for 911!
If only there was a supercomputer that could revise news posts before they go live? It could be in the form of *gasp* an editor!?
"i'm sorry .... so sorry. it had to be done."
No it didn't.
What are you, 10?
- It's not the Macs I hate. It's Digg users. -
That much heat in one place has got to wake up something doesn't it?
- Adam L. Beberg - The Cosm Project - http://www.mithral.com/
Big deal!!!
I only have to overclock my Pentium 4 83000 times to beat that little pocket calculator.
(Pentium 4 3.06 GHz has a theoretical max of 12 Gigaflops)
The Internet is full. Go Away!!!
That's when members of People for the Ethical Treatment of Animals are impotent, right?
Give a man a fish and he'll eat for a day. Teach him to fish and he'll wipe out the species.
No animals will be harmed in the production of this computer.
Superconducting supercomputer. Too expensive but maybe need to build one to see how they work.m .htm
http://www.hq.nasa.gov/hpcc/insights/vol6/superco
Using 'general' processors is cheap but the wrong direction according to the best supercomputer expert from Stanford. He designed some cray computers.
http://content.techweb.com/wire/26802955
"The supercomputer has a performing capacity equivalent to 500,000 high-functioning computers, the business daily said." Forbes, June 22, 2005 Haha ha.
Now if only we could find 500,000 high-functioning computer users who didn't magically have the unerring ability to turn a machine that powerful into a sluggish virus infested mess in five minutes flat of free pr0n surfing. Now that would be an accomplishment.
If my grammar and spelling are off, I am [distracted/tired/careless] (take your pick)
But I want to build a 20 PetaFlop computer. Take that Japan!
Yeah, that's what quantum computers will do.
Fun with Anagarams! LADS HOST, SHALT DOS. HAS DOLTS. AD SLOTHS, HATS SOLD. ASS HO, LTD.
resistance to spilt liquids?
I guess you could say they're Peta-philes.
How many BogoMips is that?
First, it seems almost powerful enough that it might start and run Adobe Premiere within four or five hours instead of six or seven.
Second, Kingdom of Loathing would finally have zero lag on the server side.
Third, it might be slightly more resistant to Slashdoting and building a router out of one of these might complete the defense.
Fourth, by the time this ends up on my desktop, Duke Nukem Forever will be in beta.
Other than that, should make wonderful blurb filler regarding chess matches with Russians for kids' science news periodicals.
If my grammar and spelling are off, I am [distracted/tired/careless] (take your pick)
Yeah, but as it goes onto say in that article there's currently not enough $'s in making HPC chips for anyone to find the job commercially worthwhile.
:)
It'll be an interesting decision for the Japanese. The Earth supercomputer used custom NEC processors IIRC. It was a top performer for quite a while. The other current top performers are PowerPC, Itanium and Opterons. Maybe not vanilla processors but not custom either.
If you remove commerical pressures then maybe custom makes sense... we'll see I s'pose
Cheers
Stevo
Forget the truth. Science is fact.
Bugger when your tea flops. Especially a high tea.
The wording of the article is terrible. "Japan Wants to Build 10 Petaflop Supercomputer", I want to build a 10 petaflop computer, too, does that mean I am capable of that? No. The difference is Japan set forth the process of creating a 10 petaflop computer. The article should read something like "Japan Building 10 Petaflop Supercomputer".
and besides we were talking about tea. Where do you get ideas about coffee from a talk about tea?
Why was that modded offtopic? A lot of us are idiots and we do want to know.
Eh.
My question was wouldn't the result be the same, not whether or not it would be scalded if pouring it into boiling water. That is, if it would get scalded by pouring it into boiling water, wouldn't it get scalded by pouring hot water into it?
I don't really know why we love gigantic computers, though. I live in a prefecture which is Japan's answer to rural Iowa and we built a 1,300 node distributed supercomputer without any idea of a feasible application to run on it -- we ended up computing a few zillion solutions to N-Queens before mothballing the project (I was hoping for enough CPU time to take the world record back from the real supercomputer at the Japanese university that currently holds it, but unfortunately it was not to be).
Help poke pirates in the eyepatch, arr.
What are you, new here? Of course it had to be done! It was however already done in previous posts I believe. GP was half right.
I am Spartacus
Why don't they ever mention the real world stats or operational supercomputers? /C which is falling behind, but at least they're not reporting any numbers until it actually works.
They keep saying BlueGene/L when it's not even completed (maybe it finally is). There's also
The fastest operational (like anything else matters) supercomputer is Columbia at NASA. And guess what? It's doing a ton of usefull work, like helping make sure the Space Shuttle launches without a hitch by computing all the Thermal Protection System problems and various other analyses.
Look at the number of processors it uses and it's performance compared to the others. It's one of the more efficient of the bunch.
Just wait until they upgrade it..
Top500 should include different rankings, like efficiency or measurable areas other than projected TFlops. In the end it's not how many you got, but how well you can use them.
-- Robi
That's almost as powerful as the PS3 will be! Amazing.
so should it still rhyme with beta, zeta, eta, and theta? Or should it be pronounced like pita bread?
Liberals call everyone Nazis yet they are the closest thing to it.
From what I recall about Peloton(that's what the presenter called it), they wish to have a 14.8 TF/s scalable unit with 4x Infiniband interconnect. This scalable unit itself is more than half the power of Thunder(ranked 7 in Top 500) http://top500.org/lists/plists.php?Y=2005&M=06 They plan to have 16 such scalable units.
For those who are interested in the specs: Peloton is 16 SU with 236.5 TeraFLOP/s, 215 TiB memory, 5.0 PB global disk system with 6,720 SMPs and 48+24 = 72 IBA 4x DDR sw. Power is 4.05 MW.
It's probably short out Duluth.
MSBPodcast.com The opinions expressed here are my own. If you don't like 'em... Think up your own stuff.
A computer that fast could nearly think!
Modern copyright is theft of culture from everyone and it retards the progress of the useful arts and sciences.
So this new Japanese supercomputer is running at a whopping 10 brainsecs!!! Imagine, you could simulate about 9 people or 47 slashdotters in that supercomputer (some of the power would be required to manage the simulatioins).
Seriously though, AI research will go mainstream with the first supercomputer that can process at greater than 1 brainsec.
It is your personal duty to fight for what is right on a daily basis. Ignoring injustice is identical to approving
I thought the fastest supercomputer was the flux capacitor. You know, at the calculated moment, you start off from down the street driving toward the cable accelerating to 88 miles per hour. According to the flyer, at 10.04pm lightning will strike the clocktower sending 1.21 gigawatts into the flux-capacitor, sending you back to 1985.
READY.
PRINT ""+-0
A computer with a 500 teaflop capacity is able to produce a perfect cup of tea with just a hint of honey and milk in it in under 2 minutes, while still evading a Vogon fleet.
Unfortunately, it cannot run Linux.
It is your personal duty to fight for what is right on a daily basis. Ignoring injustice is identical to approving
According to Sony's published specs, the PlayStation 3 will top that number easily!
If you were to find yourself standing on the golden gate bridge, would you think "Well, somebody is going to jump, so I might as well do it."?
paintball
I honestly dont' think AI research will get big until we discover how to recreate human intelligence. The reason why we don't have true AI today is because we don't have an HI (human intelligence) algorithm in existence - not because our computers are too slow. Even if we suddenly could build cheap computers running at 1000 brainsecs, we wouldn't be able to make them think or make human-like decisions, imo.
:)
I find strong AI - the study of computers that can think like humans - to be a fascinating subject. I believe when (if?) we discover how to do it, a huge revolution in computing and the human existence in general will occur. Every single aspect of our lives could be improved by automatic thinking machines. Hopefully it would be more enjoyable than the matrix and terminator movies would have us believe
HeHee, I wish I could break in this machine once it exists, and somehow get it do some some F@H for me :D
Well, a petaflop may be a brain-second (assuming the article is correct), but a petaflops is one petaflop per second = 1 brain-second / second = 1 brain.
I'm not quite comfortable measuring computing power in brains.
But in the US they have real goals with most of the super computers. Nuclear weapons research/testing is a popular one. Apparantly supercomputers are good enough these days to actually test current stockpiles via simulation. This is useful, given that the US is a signatory on a nuclear test-ban, which applies to actual detonations only (you can screw aroudn on camputers all you want, just no actual blowing up of weapons).
Weather modeling is another favourite.
As I understand it, all the Blue Gene series are targeted at organic modeling of one kind or another, protein folding, DNA research, etc.
Now maybe they don't NEED to be built, in the strict sense, but it seems they are just built for no reason. They have specific goals of what they want to try and do with them when they start work on them.
Not that the government doesn't do things purely to keep skills alive, that's more or less the reason for the US's new attack subs (Virgina class), and it's even been stated as such. They aren't building them very fast, since there's not a real need, but it's useful to slowly modernize the fleet, and you want to keep the industry around and skilled up, so you don't go through a crisis if you suddenly need to get more built.
However supercomputers do seem to have specific tasks in mind when they are built, they don't just seem to be being built to have big computers around.
Based on this, admittedly rough set of criteria, we can make a prediction. If the japanese 10 brainsec computer is deployed in 2010, then by 2020 such power should have drifted down to Universities. That works out to sentient software by 2030. Funny how it is always 20-30 years out...
Perhaps by 2050, such a beast might fit into a box the size of a modern PC, but it is going to require quite a revolution in fabrication (probably nanotech construction). Ultimately, it should be possible to see real robots in 40-50 years (and big honken thinking machines in less time).
It is your personal duty to fight for what is right on a daily basis. Ignoring injustice is identical to approving
I propose that the submissions will be done in leet speak, this would make the spelling mistakes less noticeable .
I bet you wouldn't have been modded down if you hadn't apologised...
I doubt that will advance AI research as far as people would hope. We still don't really have any good AI algorithms and even with that much computing power brute force AI isn't going to work well. From what I understand the brain works so well becuase it works efficiently, its like the difference between a nuclear reactor and a Nuclear weapon. The nuclear weapon is more powerful, but the nuclear reactor can be used to do things more effectively.
Although F-2 cost a lot more then the F-16 will ever be, it did saved a lot of money on the initial R&D by adapting a good design, and F-2 is not a simply enlarged version of F-16, it did has a lot of innovation and locally designed components goes too.
Japanese's military traditionally try to build everything themselves either indigenous design or licensed design, and doing so they achieve technology independence, at the cost of hugely inflated cost.
It's the same for super computers too.
The butter pot is sold separately.
OS Software is like love: The best way to make it grow is to give it away.
Nowadays the supercomputer contest is just a matter of who can buy the most Opteron PC's and Cisco routers from Newegg and connect them. You might as well buy a few million DVD's from Best Buy and say you have the world's largest hard drive.
Eventually small countries will connect all the computers of their entire population with distributed clients and call that the world's largest supercomputer.
This business of entering a command, waiting a minute for zillions of nodes across a slow network to start, and waiting another minute for all the nodes to finish is hardly what supercomputing used to be.
It would be more interesting to see who does the most work with the least latency or who does the most work with the simplest programming model. Anyone can write a massively parallel program to utilize every Opteron in the world but a computer which can do the same work sequentially seems like a much bigger step forward.
They'll probably turn it into some super hentai-server.
The problem is not the hardware.
It's the software.
If you could create a consioucness with an algorithm, then you could run it on a regular PC - it would still be 'aware' but would just perceive time on a much slower scale than we do due to it running so slow.
FIPS like MIPS is meaningless as alot of it is dependent on how many intructions it takes to calculate a result. If the CPUs are RISC based complex operations will take considerably longer than the same setup with CISC (and as for patern searching both would be slower than a similar speced ZISC set up!)
In the not too distant future, next Sunday A.D.
Well, a flop is a failure (a petaflop is actually a very big failure; I guess the Japanese actually are aiming at a 10 petaflop s supercomputer; FLOPS is short for FLoating point Operations Per Second). Therefore a "Metaflop" probably is a failure to fail.
The Tao of math: The numbers you can count are not the real numbers.
Well, it actually worked quite fine. The AI was so intelligent that it discovered by pure thought that the only way to keep the humans to annoy it was to pretend it would not work well. While this of course meant that the AI would soon not be run anymore, that wasn't a problem for the AI, since it didn't fear inexistance.
The Tao of math: The numbers you can count are not the real numbers.
It's a bit early for PlayStation 4 hype isn't it?
A pizza of radius z and thickness a has a volume of pi z z a
Actually it's only able to produce a liquid which is almost, but not quite, entirely unlike tea. That's why it's a teaflop: A failure to make tea.
The Tao of math: The numbers you can count are not the real numbers.
Seriously though, AI research will go mainstream with the first supercomputer that can process at greater than 1 brainsec.
No! AI research will go mainstream the moment the first desktop workstation machine passes 1 brainsec. Cos who's going to waste $10 million on using a cutting-edge supercomputer to do completely blue-sky crazy research, when there's nukes and hurricanes and proteins to simulate?
That is classic! I was going to reply with a similar comment about the runner beans giving the Lima beans a chance...
We have a long, long way to go to come up with real AI.
All AIs we have ever worked with in our history have been what are called heuristic AIs--they have a specific program which specifies how to learn, and they do so. However, there is no way they can break the bounds of their program, and they only do exactly what we program them to do. In the end, this is not very useful.
The next revolution in AI will be when someone figures out how conciousness in the brain works, figure out how to imitate it on a computer, and finally use it to create a truly intelligent AI, not just a bunch of heuristic scripts.
The final revolution will be when the first self aware AI is created.
This may happen in 5 years, it may happen in 500. This is because it has nothing to do with our processing capability--its a matter of theory, not of raw power.
Imagine a Beowulf clu....
Sorry.
It is your personal duty to fight for what is right on a daily basis. Ignoring injustice is identical to approving
It's time for kicking ass and chewing gum.
You have 0 gums in your inventory.
Exits are North, South, Down.
>_
I think you would be surprised of how much cpu cycles are effectively used for just keeping your computer and operating system running. But on the other hand I agree that brains and computers are totally different, but at another point:
flop = floating point operation = multiplying two numbers with comma's (ok ok, exponents). This takes me way more than one second to do, so I have something like, 0.1 flops capacity?
I can however in split seconds render depth maps when I see a room, recognize situations, objects, persons, sounds, perform highly complex sequential tasks (think for a moment about an algorithm for a robot to just dress himself). This is where the petaflops comparison holds. The reason why we are able to do this is because our brain is massively parallel: 50 billion interconnected (very simple) chemical processors, who communicate with a scala of electrical and chemical signals.
int main(void) {while(1) fork(); return 0;}
measure supercomputers in logic inferences, not in floating point operations. We can hardly build an AI from nuclear simulations or weather maps.
There you are, staring at me again.
Am I the only one who read the headline and thought, "Why the hell are the japanese wanting 10 supercomputers?"
I just figured they wanted to hold the world's fastest LAN party.
I8-D
Wow, this list moves fast. Last time I checked (admittedly about a year ago), the earth simulator was flogging everyone.
You would really be on to something :)
So at 10 petaflops, it will only take 3.5 months of constant training before the computer keeps asking why to every response you give it? A year and a half before it can learn to drive a car? Speaking of which, is there a possibility that - given enough training - it can lear to talk on the phone and drive at the same time? Now THAT would be superhuman processing.
Is it just my observation, or are there way too many stupid people in the world?
The next revolution in AI will be when someone figures out how conciousness in the brain works, figure out how to imitate it on a computer, and finally use it to create a truly intelligent AI, not just a bunch of heuristic scripts.
The final revolution will be when the first self aware AI is created.
I really hope that doesn't happen soon. I don't care that AIs could take over the world. I'm thinking that humans would create human level AIs for slavery to do the one task that humans want done and try not to think about anything else. That would encourage AIs to do away with humans or atleast leave us behind and go else where. (When AIs could pick up and leave it would cause a global problem and civilation would be knocked back by a few centuries. (That's if we are lucky.)
Of course it could be worse. We be playing with "transfering" human thought to computers and some idiot does it, but the process isn't perfect and the result is an insane human in an AI form. Depending on how linked the world is, then alot of damage could be done. We really need to create some sort of civilization renewal college where all the basic tech to get back to where we currently are at is taught. Most humans don't tend to think that long term though.
"Current fastest [publically ackknowledged] supercomputer is the partially finished Blue Gene is capable of 136.8 teaflops and the target when finished is 360 teraflops."
10 Petaflops? Why, that's barely half what Sony has claimed the PS3 will be capable of...
Petaflops is so early 21st century. Got to talk in exaflops now.
I havent seen this lame joke in while.
True - we won't be able to test AI theories until the computer is powerful enough to test something on. But I think an even bigger problem and a major bottleneck is that we have no clue how the brain does some very basic things. For example, computer vision is a huge problem; if you gave a room full of the 10 most brilliant vision researchers 10 computers of 100 brainsecs each, I don't think they would know how to recreate object matching in a method like the brain. You can show a 4 year old child a beat up 1985 chevy, tell him its called a car, and then ask him what a bmw is called. He will likely tell you its a car. This is called invariance, something that we still don't entirely understand how to recreate in a computer. I bet you those vision scientists will have a hard time programming something that can learn what a car is.
Do these fastest super computer claims take into account grid or other distributed computing? An example that comes to mind is the computers running SETI@home could be thought of as a SETI@home super computer. It would only be fair to include such "virtual" super computers for a true comparison.
How about microprocessors?
The road to hell is paved with good intentions.
The next revolution in AI will be when someone figures out how conciousness in the brain works, figure out how to imitate it on a computer, and finally use it to create a truly intelligent AI, not just a bunch of heuristic scripts.
Not necessary. AI doesn't have to internally work same way as human brain. As long as it simulates a task close enough, that's good. A plane doesn't fly the way birds fly, and yet it's good at what it does. If AI would work the way brain works, we could as well drop the "A" from AI, couldn't we?
The purpose of life is to find the purpose of life.
The reason that computer vision isn't a reality is that everyone has approached it the wrong damned way. I am also an AI researcher and I have the solution to computer vision. The answer is computer imagination.
Think of it as a guessing game (and please watch yourself do the same thing). Imagine that there is a feedback loop within the Computer Vision AI. First, it takes the images from it's sensors and does a first-pass using traditional techniques. Second, it uses that first pass to provide hints to form a 3D computer model of what it is seeing. It renders the model and compares against what it sees. From the point of first comparison, we must use a combination of hill-climbing and a noisy search (such as a constrained genetic algorithm) to attempt to identify models to be included and refine the 3D model that represents the current scene. The goal is to get the rendered scene to match the perceived scene by making informed guesses and refining the model. The result, after enough iterations, is a reasonably accurate semantic representation of what is being perceived. Like human vision, this type of computer vision will never be 100% accurate (unless the models in the database are complete and the system is provided with large numbers of brain-sec cycles to figure the complete scene out), but it doesn't need to be 100% accurate to correctly analyze the most important elements in it's vision.
Humans do something very similar to this, and our feedback mechanism includes very complex analysis (Is that Joe? I can't see his full profile, but that looks like his nose.. No... That's not Joe, his chin is pointier... Maybe that's his brother?)
For AI, always think IMAGINATION and COMPARISON, FEEDBACK, EMERGENT COMPLEXITY and FRACTAL PROCESSING (Meaning that processing networks are combined to form more powerful processing networks in combinatorial patterns that are self-similar at different zooms).
It is your personal duty to fight for what is right on a daily basis. Ignoring injustice is identical to approving