Fotolia
Iran says it's building an AI supercomputer, despite sanctions
In a tweet, Iran said it is building a new supercomputer. Other nations are doing the same because high performance computing platforms lend themselves to AI development.
Iran says it is planning to build an AI supercomputer and will have it in operation sometime next year. It will have to work around U.S. sanctions to get the tech it needs, but it has succeeded in doing so in the past.
The news came in a tweet last weekend by Mohammad-Javad Azari Jahromi, Iran's minister of information and communication technology. It was in Persian, but translated by Google, Jahromi said the new supercomputer, "is due to launch next year: 100 times more powerful than previous ones, great!"
This supercomputer, wrote Jahromi, "will serve people to support businesses with the goal of developing artificial intelligence." The system is named "Simorgh" after a mythical bird.
ابررایانه ایرانی "سیمرغ" تا سال آینده آماده میشود: صد برابر قدرتمندتر از نمونههای قبلی، عالی و فوقالعاده!
— MJ Azari Jahromi (@azarijahromi) August 17, 2019
این ابررایانه برای حمایت از کسبوکارها با هدف توسعه هوشمصنوعی به مردم خدمت خواهد کرد.
هزار آفرین به سازندگان جوان و خلاق. بچهها ممنونیم!
#آیندهروشن✨
Translated from Persian by Google:
The Simorgh Iranian supercomputer is due to launch next year: 100 times more powerful than previous ones, great! This supercomputer will serve people to support businesses with the goal of developing artificial intelligence. Thousands of happy young and creative creators. Thanks guys!
Iran isn't doing anything different from other nations in building a system designed for AI data intensive workloads. Take France, for instance. It is building a supercomputer to accelerate development of its AI industries. The French system, due in October, may rank on the Top500 list as one of the 20 most powerful supercomputers in the world. Since 1993, the Top500 list has ranked the world's 500 most powerful supercomputers twice a year.
However, the difference between France and Iran in how the systems are built is considerable. Hewlett Packard Enterprise is building the French system. Iran, if it wants U.S. processors, may have to work through the black market because of U.S. sanctions.
Trade restrictions have not prevented Iran from acquiring what it needs. In 2007, for instance, Iran's Amirkabir University of Technology announced it had assembled a Linux-based system using 216 AMD Opteron cores. It even published photographs, since removed, of the system under assembly. Iran has announced supercomputer efforts in 2011 and 2014. It's unclear what Jahromi is benchmarking against when he characterized the forthcoming system as "100 times more powerful."
HPC is ideal for AI development
HPC systems are seeing increased use for AI development because they "are very adept at handling massive volumes of data," said Steve Conway, senior vice president of research at Hyperion Research.
HPC systems use advanced parallel processing, have more internal storage and memory technologies, and can move data around much faster than standard enterprise servers, Conway said. "All of that contributes to being able to solve problems not only much more quickly, but handle bigger problems all together," he said.
Hyperion recently surveyed 54 thought leaders in HPC on the "importance of HPC for advancing AI." Nearly 90% said it was "extremely important."
Although Iran can use non-U.S. chips, such as China's homegrown processors, Conway suspects it will want U.S. tech. "Chinese chips are certainly capable, but there is not a ton of software yet written for them," he said.
Iran has never had a system on the Top500 list, and there's no apparent independent verification of its capabilities. But Iran has clear scientific and engineering capability. In 2016, it ranked 15th globally in a National Science Foundation report that looked at scientific and engineering articles published in peer-reviewed journals, books and conference proceedings in all fields by country. That said, the rankings, overall, are an imprecise measure because they don't consider the quality of the work.
