Slashdot Mirror
India's Bargain Supercomputer
MaximusTheGreat writes "India beat U.S. supercomputer sanctions by building a teraflop $5 million PARAM Padma supercomputer, which is half the price of similar computers being sold in the international market. It can be scaled upto 16 teraflops, on a build-to-order basis For comparison, the fastest supercomputer in the U.S. is about 10 Teraflops. Some techical details and more info on CDAC , ITworld, Economic times and Asia Times. Also, India has been exporting older model PARAM 10000s to other countries like Russia, Canada, Germany etc. for some time, and expects to increase exports significantly with the new model PARAM Padma."
The article says it was built with ibm power4 chips operating at 1ghz. What I am wondering is this - if the US gov't decides to get upset about this will they prohibit IBM from selling power4 chips to the company that makes these supercomputers?
And of those of you interested in the power 4 check out this page. A pretty cool chip from what I can tell...
So those fiendish Indians won't be able to simulate nuclear warhead tests, that's why. If they could do that, they could BUILD a working one...errr....no, wait...If they can't simulate, the only way to test improvements in warhead yield is by detonating a warhead for real, and that's better because....hmmmm...
Ya got me. Why don't we want India to have a supercomputer?
If a job's not worth doing, it's not worth doing right.
It's amazing what pressure can do for you. Until a few years, we were denied technology for building supercomputers by the west. This forced CDAC to work on such technology from scratch.
Likewise for most of India's rocket programme (albeit with the occasional help from Russia) and other technology. When pushed to the limits, you outperform yourself.
I'm sorry, but I have to insist that you can't add a bunch of PC's together, and end up with the equivalent of a general purpose supercomputer.
These so-called "supercomputers" only have the performance of a supercomputer on the class of problems which are inherently parallelizable. For problems which require serialization, by needing results in hand to go onto the next step in the calculation, these things slow down to the speed of one of the component PC's.
I guess now that Seymore Cray has died, no one else can build real supercomputers. 8-(.
-- Terry
And India IS nuclear and has a fairly stable government. The U.S. or U.N. will be hard pressed to tell them "bad". Plus I really wouldn't mind their nuclear program going digital, a little less radiation in the morning breeze. They're going to do nuke research no matter what, might as well encourage them to not blow shit up in the process.
To top it off, why is it exactly we, as THE U.S. feel we can be top dawg and keep everyone else down. India, like the U.S., has quite a bit of poverty and it really wouldn't hurt to encourage them to look towards bettering themselves and their people. (nuclear research can actually be used for things other than building bombs ...)
In the tech world there are no borders, and this really pisses the old schoolers off. We're not going to do anything except identify they broke a sanction trust me we've picked enough fights to last us a while ... plus just think about the benchmarks they'll get on that baby ... it will just kick the shit outta everyone else at the U.N. LAN parties ...
Ignore the "p2p is theft" trolls, they're just uninformed
Its better because its really hard to hide a detonation, for obvious reasons. While this is certainly not an ideal situation, at least it is supposed to ensure that the world knows who the nuclear powers are. Using computer modeling, a country could develop and test a nuclear weapon without anyone knowing. I think the idea is that its better to blow up some desert scrub land, than a major city as the first indication that the country or group has nuclear weapons.
Degaussing scares the bad magnetism out of the monitor and fills it with good karma.
Seems to have been in the news lately. From Bill Gates to hand held computers - it seems to have seen a lot lately. A short summary -
(1) Bill Gates and India
The AIDS donations, and yah, maybe source code sharing and Windows for a bargain.
(2) The Simputer A handheld pc that runs on Linux and is available for a bargain.
(3) Reliance and CDMA
Reliance has launched a WLL based mobile phone system which promises high bandwidth, JAVA enabled advanced telephony. AT pathetically low prices !!!!
(4) The Param Padma
A 16 teraflop supercomputer, faster than many others available.
Coming to think of it, a developing nation seems to have more to look forward to than a developed nation!! That also reminds me that the brain behind many Intel chips is Indian. India lacks infrastructure/proper governance/organization. But it looks like Nostradamus's prophecy of India becoming a superpower may come true... within the next few years.
|/________
|\A|ALYS|
These so-called "supercomputers" only have the performance of a supercomputer on the class of problems which are inherently parallelizable.
:-)
Sure, but that was also true in Seymore Cray's day. Admittedly his machines provided substantially better scalar performance than normal mainframes did at the time, but their rated "supercomputer" performance was only obtainable by operating on multiple datasets simultaneously using vector operations. If your problem wasn't parallelizable, your investment was largely wasted. Fortunately, there is no lack of real-world problems that are inherently parallel.
It's worth noting also that conventional single-machine performance is inherently limited by a number of very strong physical constraints, primarily the speed of light and device leakage currents as transistor geometries become ever smaller. This really leaves us nowhere to go for pure sequential computation speedup. Quantum computers will probably live in a single box to reduce decoherence, but even they require problem breakdown into parallel solutions.
Hence the move away from scalar performance and towards multiprocessor, cluster, and distributed computing. With SETI@home delivering around 15 Teraflops on their specific problem at the distributed end of the spectrum, and with the PARAM Padma's IBM Power4 providing 2 processors per chip and 8 processors per module directly out of the IBM fabs even before assembly of multiprocessors and clusters commences, clearly computation is heading towards a future of high parallelism. And about time.
"The question of whether machines can think is no more interesting than [] whether submarines can swim" - Dijkstra
Islamic fundamentalism is one good reason. Other then that, we have history: It is Pakistan that has always attacked India. Three times, methinks, although I'll let other supply the years.
There current "ruler" himself attacked even when their PM was holding peace talks.
Pakistan is a military/Islamic dictatorship, one religion; India a democracy, cultural meltpot.
Successive Pakistani regimes have shown they have no qualms about speaking lies and twisting the truth as long as it serves the purpose of fueling the hate against India. The Art of War: In an extremist enviornment, the more extreme or radical your position, the more powerful you become. Same happened in the Talibanesque Islam. Many other examples too.
Pakistan is a nation which, IMHO, has it's identity derived from one fact alone : enemy of India. They have hardly any progress to cheer about, even when you compare to India, and that is saying something.
Large parts of Pakistan's NWFP are actually not exactly under administrative control of the government. Fundamentalism/tribals rule.
Pakistan is the haven for criminal activities. Daewood Ebrahim, belived to be one of the richest persons in the world, operates from Pakistan, through ISI support. Islamic terrorists, in NWFP, who hijacked Indian Airline aircraft were recently discharged from house arrest!
Pakistan's armed forces can't be trusted. Their generals keep popping off democratically elected governments. They also eliminate their own Generals. Figures. Large section of Army commanders and ISI are sympathetic to the "Islamic cause". Imagine, in such a situation, what a commander with nuke trigger might do. He'll launch the "Islamic bomb" against India to ensure martyrdom for himself and his "religion".
And fuck up millions of people. That's why Pakistan is more likely to "lose it" than India.
If the USA hasn't attacked Pakistan like Afghanistan, it is because of some reasons:
1. We need(ed) a base to carry out operation in Afghanistan against Taliban & Co.
2. Fear of Nuclear holocaust.
3. The fact that Pakistan is the centre of Islamic terrorism. Smartly enough, the Bush administration is actually using the military dictator/self appointed president to track down and destroy Al-Qaeda network.
Or that, atleast, is the public line.
Certainly, we have managed to nab a few top Al-Qaeda operatives in Pakistan. Reports keep coming how Osama is only a grasp away. In Feburary, he was almost nabbed; ran away just an hour before a joint raid by US intelligence and ISI. Go figure.
Of course, we are not pressuring Pakistan to dismantle the Islamic terrorist organizations that operate in India. Not unless this business of Al-Qaeda is done with. There are disturbing reports that Al-Qaeda is regrouping in Karachi (the crime capital of the World) and planning to "utilize" the emotions once US attacks Iraq to launch more attacks against US/India/Israeli targets. Let's see.
Most technology really only requires a bit of drive and a few smart people to develop. Hell, that's where it came from in the first place.
Most of America's restrictions on exporting tech overtly rely on the concept that America is just better and smarter than anyone else.
Bull puckies.
If America wants to maintain any sort of commercial lead in technology it has to distribute it in such a way that it's just plain easier and cheaper to *buy* it than develop it yourself.
As Goethe noted, everything has been thought of, the trick is to think of it again. The historical evidence is clear that anything America can think of so can China, Russia, England, Germany, etc.
Oh. Wait, as often as not these countries think of things *first* and America has to play catch-up.
The idea that you can 'restrict' technology is just plain doofey. If you can figure something out so can hundreds of thousands of others.
So go for it India. Think of stuff, build stuff, develop 'home grown' free software, put the screws to American 'tech' companies and make its government sweat bullets.
Maybe it'll get the country off its ass again, like Russia did when it launched a satellite years before anyone thought it would be possible.
Anything but this damned brand name pushing, marketroid driven 'economy' we've got now.
KFG