A trillion operations per second

ISTTHPC 2000 supercomputer with peak performance of 3.5 and 10.5 teraflops (1 teraflop = 1 trillion operations per second = 1000 billion operations per second) is developed and tested at the Institute for Space Technique and Technology and the National Academy of Engineering of the Republic of Kazakhstan.

This is a graphic chips-based device powered by NVIDIA CUDA. The test of the supercomputer assembled by experts showed performance of 2.74 teraflops. Power consumption totalled 1.71 gigaflops per watt. In the future it is planned to organize a pilot production of own supercomputers range, personal through cluster high-performance hybrid type systems based on the combination of central and graphics processing units (GPU). Experts note that the supercomputer would be used, inter alia, in modelling of oil and gas fields. 

In general, the main purpose of such supercomputers is to give huge performance for engineers and scientists work. They are applicable in a variety of industries - from weather forecasting and medicine to calculating military vehicles components. The most glaring example is that the Boeing managed to save almost 2 billion dollars thanks to the programming using a supercomputer! And reason of this is the fact that most of the preliminary tests were carried out with a supercomputer, not in air tunnels, what reduced the time of creating new aircraft for one year. As we can see, supercomputers are compensated, however without expensive software they are just useless things.

The most interesting thing is that China is a leader in production of such machines. Chinese supercomputer Tianhe-2 (the Milky Way-2) topped the regular Top 500 supercomputers rating. Test of Linpack Tianhe-2 showed performance of 33.86 petaflops (1,000 teraflops!) or 33.86 quadrillion floating point operations per second. Therefore, the Chinese supercomputer’s performance is almost twice more than such of the American Titan system. Its results were 17.59 petaflops.

0
No votes yet

Ask the author