Oxford's oldest student newspaper

Independent since 1920

5 minute tute: Supercomputers

How powerful are supercomputers?

The term ‘supercomputer’ has been used since the 1960s to describe the fastest and most powerful computers in the world. They are used for the most challenging computational problems: for example weather forecasting and climate research and simulating the formation of stars and galaxies. Computer speed is measured in ‘Flops’. This means the number of ‘floating point operations’ per second (eg. one addition or one multiplication). Modern laptops and PCs are rated around 10 billion Flops or 10 GigaFlops. The fastest supercomputer at present is more than a hundred thousand times faster- nearly 2 quadrillion Flops or 2 PetaFlops.

How have they changed over the years?

The earliest supercomputers had a single processing unit. In the 1970s ‘vector’ supercomputers were developed which had processors that could start on the next multiplication before the previous one had finished, thus speeding up the calculation. In the 1980s most supercomputers consisted of a modest number of vector processors. The nature of supercomputers changed drastically in the 1990s with a move to ‘parallel’ architectures. Parallel computers have large numbers of processing units which all work together. The earliest parallel supercomputers had hundreds of processors but the largest machines now contain hundreds of thousands of processors.

What are supercomputers used for?

In some cases computer simulations using supercomputers can provide information that cannot be provided through experiments, or a supercomputer is either safer or cheaper than performing experiments. One example is car crash testing. Before supercomputers began to be used, it cost roughly £300,000 to perform a single car crash test. If these tests were not done and a new model coming off the production line failed the legal crash tests the car would have to be redesigned at vast cost. The cost of each car crash test was almost dominated by the cost of hand building each car to be tested. Once accurate simulations of car crash tests could be performed on supercomputers, which happened around 1990, car companies were able to save millions by performing virtual car crash tests for each new model.

What’s the future for supercomputers?

In the last few years, a number of supercomputers have been built using GPGPUs (general purpose graphical processing units which are based on the technology used in graphics cards) and it will be interesting to see whether this technology dominates the Top500 lists in future years. There are concerns about the energy consumption of supercomputers and about our ability to develop software that allows all the individual processors to work together on a single problem. There are plans to build an ExaFlops supercomputer (a million, million, million Flops) by 2018.

Check out our other content

Most Popular Articles