India’s plans to try and develop by 2017 a supercomputer much faster than any available today are unrealistic and impractical because of technological constraints that have hobbled similar efforts the world over, global and Indian experts have cautioned.
, these experts – including senior officers of the very government agency tasked with building the supercomputer – have pointed to the challenges of building such a machine anytime in the near future.
A single supercomputer with the speeds proposed by the government will need to be wired to a dedicated nuclear plant as its power supply and will consume the electricity used by over 2 million Indians or 5 lakh homes. That’s apart from the power required to cool it.
And this is a best case scenario.
“It will not be possible to build an exaflop speed supercomputer by 2017,” said Pradeep K Sinha, director of High Performance Computing (HPC) at the Centre for Development of Advanced Technologies (C-DAC), referring to computing speeds the government project aims to reach. “I myself will say that.”
“But we can and must work on research towards developing such supercomputers,” Sinha said.
Telecom minister Kapil Sibal had last week written to Prime Minister Manmohan Singh seeking Rs. 4700 crore in the 12th Five Year Plan (2012-17) for a project to build petaflop and exaflop speed supercomputers. A petaflop is a measure of computing speeds and an exaflop is 1000 petaflops. Sibal suggested, in his letter to the PM, that C-DAC be asked to build the new supercomputer. C-DAC had built India’s first supercomputer, the PARAM 8000 in 1991, and has since built several more advanced versions. But India’s fastest supercomputer at present has a maximum speed of just 0.3 petaflops – 3000 times less than an exaflop – and ranks 58 among the world’s fastest machines. The world's fastest supercomputer is IBM's Sequoia which has a maximum speed of 16.32 petaflops.
Supercomputers, with their ultra fast processing speeds – equivalent to the combined speeds of thousands of PCs --help in key strategic projects including the N-program, defense projects, and tsunami and cyclone alert systems.
Like India, other major countries – like the US, Germany, China, Japan, the UK, France and Italy -- also have supercomputing programmes, and have been eying the prospect of exaflop speeds for the past few years.
In 2007, the US Defense Advanced Research Projects Agency (DARPA) formed a group of experts to evaluate what it would take to build an exaflop speed supercomputer by 2015.
The group concluded that “the practical exaflops-class supercomputer DARPA was hoping for just wasn't going to be attainable by 2015. In fact, it might not be possible anytime in the foreseeable future,” team leader Peter Kogge wrote recently in a signed article for the journal IEEE Spectrum.
The DARPA team set out with a goal to evolve a blueprint for an exaflop speed supercomputer that would consume 20MW – the most efficient power consumption they felt could be achieved. They concluded that the supercomputer – if built – would require at least 67 MW. Subsequent analysis has independently pegged a realistic power requirement at 500MW.
But even if Indian scientists could develop a 20MW supercomputer, it would consume 1728 million units of electricity (kWh) in a year – more than the electricity consumed annually by over 2 million Indians. The per capita annual consumption of electricity in India is 780 kWh.
Indian scientists however also cautioned that though exaflop scale supercomputers appear distant at present, the only way come close to building them is by investing in research now.
“They won’t get developed suddenly, overnight,” Sinha said. “Like all the other nations across the world, we too need to invest now in research so that we can evolve a way of building petaflop and then exaflop supercomputers a few years down the line.”
World's fastest machines
1. IBM Sequoia (USA)
2. K Computer (Japan)
3. Mira (USA)
4. Liebniz (Germany)
5. Tianhe-1A (China)
58. Cluster Platform 3000 (India)
© Copyright © 2013 HT Media Limited. All Rights Reserved.