Slashdot Mirror
Iran Builds Supercomputer From Banned AMD Parts
Stony Stevenson writes with the news that, despite a ban on US PC hardware, Iranian techs have built an enormously powerful supercomputer from 216 AMD processors. The Linux-cluster machine has a 'theoretical peak performance of 860 gig-flops'. "The disclosure, made in an undated posting on [the University of] Amirkabir's Web site, brought an immediate response Monday from AMD, which said it has never authorized shipments of products either directly or indirectly to Iran or any other embargoed country."
"The Iranian supercomputer falls far behind the world's fastest computers. In November, the BlueGene/L System, jointly developed by IBM and the U.S. Department of Energy was ranked No. 1 in the world with a benchmark performance of 478.2 teraflops. A teraflop equals a trillion calculations per second."
Indeed, the article mentions at the end that it falls far behind the rest of the world. In fact, to make the Top 500 this year you had to have a supercomputer worthy of 5.9 Teraflops.
Only his tendency toward a dazed stupor prevented him from screaming aloud.
And so does Christianity and Judaism. http://en.wikipedia.org/wiki/Usury#Usury_within_religious_texts
"I think an etch-a-sketch with an ethernet port would beat IE7 in web standards compliance."
Actually the link is an Australian repost of an American media story. Here's the original (as linked in the Australian repost): http://www.informationweek.com/story/showArticle.jhtml?articleID=204800653&cid=RSSfeed_IWK_All
Just to point out your history of Iran, please don't act like the US was clean. The US ally you referred to that ruled Iran was installed by the US government after an operation by the CIA to overthrow a freely elected leader. To Iran, the US is seen as terrorists, mostly for meddling with the sovereignty (sp?) of other nations.
You are wrong.
Google the news on Iran and that latest CIA report that says Iran stopped pursuing nuclear weapons in 2003. Guess what you'll find -- the EU, France, Germany and others basically saying the U.S. intelligence is flawed and Iran is a much greater nuclear threat than that report states.
France and Germany are pushing for harsher sanctions than the U.N. ones. They want separate EU sanctions on Iran, and still call their nuclear program "a threat".
The Middle East nations all are fearful of Iran as is, and terrified of them having nuclear weapons. Arabs != Persians.
Learning HOW to think is more important than learning WHAT to think.
Hehe, last time I (and some others) checked, Iranian leader did not in fact threaten to wipe anything off the map. It turned out to be a mistranslation by a bunch of incompetent journalists.
However supercomputers have now progressed to the point that you can actually TEST a bomb all in software.
This is inaccurate.
The basic nuclear design tools are finite element modeling and Monte Carlo simulations. With larger and larger number of elements modeled, you can get more and more accurate simulations in the same timeframe, so that the model has closer and closer resemblance to experimental reality. You also need some baseline data; some of that is declassified, some can be obtained experimentally on smaller scale using neutron beams, lasers, and high explosives. But the most important data on the high efficiency yield properties, and the algorithmic optimizations allowing rapid and detailed simulations, remain classified.
Even with a supercomputer design, without an actual test, you can't be sure your extrapolations and simulations will be as good as you hope. Getting a nuclear explosion isn't the real challenge; it's making one that's efficient. (This may have been North Korea's problem; sub-kiloton yields can result if you make a mistake.) However, a good computer lets you get a better idea of the sorts design variants you want to play with before you go risking your very expensively obtained fissionables on a test explosion.
But basic work and a rough model once you have the basic materials data? Two days on the HP-49 calculator, including programming time. A 7x7x7 element model gives you numbers that will be within 10% of the final... which does translate into an order of magnitude difference in possible yield, but anything from 1 to 100 kilotons still gets attention.
//Information does not want to be free; it wants to breed.