Bangalore: Largely known in India for the supercomputer that the country was denied by the US government in the late 1980s—a move that kick-started India’s own supercomputing programme—Cray Inc. is now trying to shed its “pricey” image here as it foresees exponential growth in the high performance computing market. The supercomputer maker says that buyers in the country are looking seriously at a minimum of three petaflop computers, ones that crunch a thousand trillion calculations per second.
In June, Cray Inc., whose Jaguar Cray XT5 is the world’s fastest machine according to the latest Top500 Supercomputers ranking, formed its Indian subsidiary. Around the same time, it lost a $28 million (Rs128.8 crore today) high performance computing deal of the Indian ministry of earth sciences to International Business Machines Corp. (IBM) primarily on price.
Aggressive stance: Cray Supercomputers country manager Sudhakar Yerneni says the firm is competing for big as well as small deals. Hemant Mishra/Mint
However, with its next generation interconnect technology, code-named Gemini, to hit the market in the next few weeks, about the same time that its chip partner Advanced Micro Devices Inc. (AMD) is planning to launch the world’s first 8- and 12-core processors, Cray hopes to capture a wider user-base in India. So far, AMD’s 6-core Opteron processor is the leading processor in supercomputing, according to the Top500 list, a listing of supercomputers on the basis of processing speed.
Interconnects are highly specialized network interfaces that allow a large number of processors to work together. Most high performance computing vendors use off-the-shelf interconnects, Cray makes its own.
“Even though we have been present in India for long, having installed machines ranging from $100,000 to $4 million, we were not forceful (in the market). Now we are going to be; we are giving up our pricey tag and becoming more affordable,” says Sudhakar Yerneni, country manager, Cray Supercomputers (India) Pvt. Ltd.
Yerneni’s confidence comes partly from the fact that Cray is the only company in the world focused exclusively on supercomputers, investing one-third of its revenue on research and development and, by extension, better poised to innovate. In 2009, when some of its rivals such as Sun Microsystems Inc. and Silicon Graphics International were sold (and bought), and the global market saw a decline in sales, Cray was still profitable, says Yerneni. Besides, Cray machines are the only ones that can run so-called grand challenge problems (such as weather forecasting, climate modelling, or drug discovery that need thousands of processing elements) as well as smaller problems on the same machine.
Still, even though Cray has moved from the traditional brute-force hardware to sleeker machines, providing teraflop (a trillion calculations per second) computers that can slide under a desk, it’s cluster providers such as Hewlett-Packard, IBM, and Dell that currently rule the market in India. A cluster is a bunch of low-powered computers hooked in parallel to enhance computing capability and is catching with customers who don’t have budgets for supercomputers.
The current momentum (towards supercomputing) in education, research and government establishments is poised to pick up more steam and expand to other segments such as manufacturing and financial services, says Subram Natarajan, executive information technology consultant, systems and technology group, IBM India and South Asia. “A petaflop computing system in India would perhaps happen sooner.”
Between 2002 and 2007, the high performance computing market in India grew 130%. It dipped in 2009, but according to research firm IDC, after a modest growth in 2010 it will rebound to 9% growth, reaching $11.7 billion in 2012.
It’s also a market that has, of late, seen computer makers signing up with application providers to expand their reach. Computational Research Laboratories Ltd (CRL), a wholly owned subsidiary of Tata Sons Ltd, rents out computing cycles (or roughly, time on the supercomputer) and its experience doing this shows that that non-traditional sectors too are ready for and need high performance computing solutions.
In December, CRL signed a multi-million dollar deal with India’s first and only Force India Formula One Team Ltd to offer a computational fluid dynamics solution to help it design the next generation race cars as well as enhance the aerodynamic efficiency of existing models.
“Our revenue has grown several-fold in the last one year,” says Vipin Chaudhary, chief executive of CRL who assumed office three weeks ago. With aircraft manufacturers Piaggio Aero Industries SpA and Boeing among its customers Chaudhary, an industry-academia veteran whose last assignment was at The State University of New York in Buffalo, says the idea of desktop supercomputing is finally catching on.
As more and more businesses, from animation to gaming, drug research to energy, and manufacturing to design, seek better modelling and simulation, supercomputer makers are scrambling to offer more efficient products. Last week, IBM released its seventh generation RISC (reduced instruction set computer) processor whose one specific model is targeted at building flexible high performance computing clusters, says Natarajan.
Five years ago, there were no teraflop machines in India; today there are at least 30-40 multi-teraflop computers. “In the next five years there will be at least a 1,000 teraflop machines,” says Yerneni. With Cray machines installed at places such as Saha Institute of Nuclear Physics in Kolkata, Tata Institute of Fundamental Research in Mumbai, and National Centre for Medium Range Weather Forecast in New Delhi, the company is fiercely competing with others for big and small deals.
“For some years, government centres will spur high performance computing in India; the industry, for now, is satisfied with medium-range computing power,” says computer scientist N. Balakrishnan, associate director, Indian Institute of Science, Bangalore.
“I don’t agree,” retorts Chaudhary. For Natarajan, the industry’s high performance computing imperative is evident: “As new business models develop and evolve, enterprises will pursue options to stay competitive in their respective fields.”