Home & Office

IBM exec: China to overtake Europe in supercomputing

China's drive to develop its HPC systems should see the country make inroads into the twice-yearly Top500 list of most-powerful supercomputers, according to IBM's chief of deep computing
Written by Jack Clark, Contributor

China is poised to eclipse Europe in supercomputing as countries compete to gain ground on the US, according to IBM's head of high-performance computing.

"Within a year, there will be more Top500 systems in China than there are in Europe collectively," David Turek, IBM's vice president of deep computing, told the audience at an IEEE-USA forum in Washington, DC, on Thursday, according to Computerworld.

Turek was referring to the Top500 supercomputing list, published twice a year by US and German computing academics, which ranks the most powerful HPC systems across the world.

The most recent rankings, published in June, saw China move into the number two spot with its Nebulae system. The supercomputer delivered 1.27 petaflops per second of processing power in a Linpack benchmark, behind the USA's Jaguar at 1.75 petaflops/s. One petaflop/s is one quadrillon calculations per second.

China had 24 rigs in the most recent Top500 list, tied at fourth place with Germany. However, it trailed France (27), the UK (38) and the USA (282). China has been in the list since the second half of 2001, when it entered the rankings with three systems.

For Turek's prediction to be correct, China needs to get more than 100 systems into the list within 12 months, as the pan-European tally currently stands at 144. However, that count may increase. Turek said that Europe, Japan and other countries are making investments in supercomputing that could see them threaten the US's dominance.

"You have sovereign nations making material investments of a tremendous magnitude to basically... eat our collective lunch," Turek said.

The next stage in HPC will be exaflop (1,000 petaflop) systems, which are expected to arrive between 2018 and 2020. A Darpa report (PDF) estimated that the cost of developing and deploying the first exaflop-scale rigs will be between $1bn (£630m) and $2bn. Subsequent deployed systems will cost hundreds of millions of dollars each, it estimated.

The same report estimated that current HPC systems cost between $1m and $100m. The figures suggest that while cutting-edge HPC systems will be more powerful, they are also going to be more expensive.

A European partnership, Prace, has begun building a pan-European grid of interlinked supercomputers, whose combined processing power should break the exaflop barrier by around 2019. However, this HPC strength will not be recognised in the Top500 list, which assesses single systems only. Though immensely powerful in chorus, the Prace computers will not be heavy hitters in their own right, and it is unlikely they will rank high in the list.

Editorial standards