Thirst for raw powerhttps://indianexpress.com/article/news-archive/print/thirst-for-raw-power/

Thirst for raw power

China’s Tianhe-2 became the world’s fastest supercomputer in June,setting a new record in strategic technology. While this field grows at a phenomenal pace,the need for speedy calculations accompanies it. Supercomputers can zero in on solutions,eliminate duds,and test probabilities of variables much faster. Here’s a look at these gigantic arrays of processing power and what lies beneath

What is a supercomputer?

It comprises an array of processors plugged into one vast network. These machines are generally used for research,where performing extremely complex calculations as fast as possible is essential. A simple example is finding the highest possible prime number. It is an endless quest,and a supercomputer delivers the output much faster. Today’s supercomputers are put to work on weather forecasting,quantum physics,simulations that try to mimic the birth of the universe,among other computationally intensive tasks

The first wave

The first supercomputers were made in the 1960s. Control Data Corp (CDC) made the first supercomputer called CDC 6600,in California in 1964. IBM owned the fastest computer in the world at the time,and the CDC 6600 proved to be around 300 per cent faster. This machine was used for nuclear physics research. Like all fast computers of the time,CDC 6600 was gigantic,with 400,000 transistors and over 160 km of wiring. It held the top spot for five years,after which it was replaced by CDC 7600,which was 10 times faster than 6600. Seymour Cray was a key member in the development of both computers. In 1972,he left CDC to found his supercomputer firm called Cray Research Inc.

Speeding up

In 1976,Cray-1—an 80 MHz supercomputer,worth $10 million—was the fastest supercomputer. To put that in context,today’s cheapest smartphones (under

Rs 3,000),have a minimum processor frequency of 256 MHz. Cray Research built the world’s fastest supercomputers for over two decades,which is why Seymour Cray is called the father of supercomputers. The company’s crown jewel was Cray-2,released in 1985. In this machine,parts were aligned in a circular shape to ensure maximum speed. The hand-wired computer had eight processors and was liquid-cooled. Its peak speed was 1.9 gigaflops. This was dethroned in 1990,when computer engineers in the US and Japan helped various firms build computers with thousands of processors.

Understanding FLOPS

Advertising

Supercomputers’ speed is measured in FLOPS—Floating-point Operations Per Second. These refer to complex calculations which are used as a measure of a supercomputer’s performance.The performance of a single processor can be measured in the number of commands it can execute in a second. FLOPS often has a prefix to it,such as megaflop,gigaflop,teraflop or petaflop. These signify the speed of calculations. Today’s fastest supercomputers operate in petaflops.

Records tumble

Intel ASCI Red was the first supercomputer to perform faster than 1 teraflop. It took only 10 years to break the petaflop barrier,as Japan’s RIKEN research institute built the MDGRAPE-3. It could operate at one petaflop per second; a quadrillion calculations per second. Two years ago,one of the most well-known computers—IBM Roadrunner—was built in the US. It was the first general-purpose supercomputer to touch the petaflop mark. Today,computational records website top500.org lists China’s Tianhe-2 as the world’s fastest supercomputer. The record stands at 33.86 petaflops.

Supercomputers in India

Param 8000 was India’s first supercomputer. It was built in 1990 by the Centre for Development of Advanced Computing (CDAC) in Pune. Top500.org lists 11 Indian supercomputers among the top 500 as of now,and the fastest is Prithvi (rank 36). Among others in the list is Param Yuva II,made by CDAC in February this year. This cost Rs 16 crore and performs at 524 teraflops. ISRO has the SAGA-220,whose name represents its top speed (220 teraflops). India plans to build a machine with a top speed in exaflops (10^18 flops) within the next decade.

The final frontier

In spite of the progress made so far,the final frontier is still at least 10 years from being breached. A recent experiment in Japan tried to mimic the activity of the human brain. The report says: “The simulation involved 1.73 billion nerve cells connected by 10.4 trillion synapses and was run on Japan’s K computer; the fastest in the world in 2011. It took about 40 minutes to complete a simulation of one second of neuronal network activity in real time.” Take a minute to think about that. An array of 82,944 processors,the fourth fastest in the world,took 40 minutes to mimic one second of neural activity. No human can use 100 per cent of the brain,but each one of us can think faster than a supercomputer.

Download the Indian Express apps for iPhone, iPad or Android