With IBM's Summit supercomputer, the United States reclaims China's top spot in high-powered computing



[ad_1]

After five years of playing in China, the United States is once again allowed to boast the world's fastest supercomputer with a gigantic IBM machine called Summit at the Oak Ridge National Laboratory in Tennessee.

Summit's performance was revealed Monday on the Top500 list of supercomputers, which a group of university researchers update twice a year, with a speed score of 122.3 quintillions of mathematical operations per second , or 122 petaflops. To match this speed, every person on the planet should perform 16 million calculations per second.

The rate of improvement has slowed for the 500 fastest supercomputers in recent years. The three lines, from top to bottom, show the total power of the machines on the Top500 list, the most powerful model, and the 500th on the list.

Top500

The Summit system, with 9,216 IBM processors boosted by 27,648 Nvidia graphics chips, takes up as much room as two tennis courts and as much power as a small town. It will be used for civil research on topics such as materials, cancer, fusion energy, fusion energy, astrophysics and the changing climate of the Earth.

Another system called Sierra, a little smaller but built with the same components, will be used for nuclear weapons research at the Lawrence Livermore National Laboratory in California. He won third place with a speed of 71.6 petaflops, behind the Sunway TaihuLight of China with a speed of 93 petaflops.

Some of the racks of computer nodes that make up the Summit supercomputer built by IBM at Oak Ridge National Laboratory in Tennessee.

Some of the racks of computer nodes that make up the Summit supercomputer built by IBM at Oak Ridge National Laboratory in Tennessee.

Genevieve Martin / Oak Ridge National Laboratory

Supercomputers are prestige systems but at high cost. The US Department of Energy's Coral program for the construction of Summit and Sierra, for example, cost $ 325 million. But they have played an important role in everything from modeling nuclear weapons without real explosions and simulating the universe with a method called computational cosmology.

The work of computational cosmology allows the SLAC National Accelerator Laboratory near Stanford to understand what makes the universe vibrate. This video shows galactic collisions that occur in the real world over hundreds of millions of years.

Video by SLAC / animation by Stephen Shankland / CNET

Build a better benchmark for supercomputing

There is a national race at a performance threshold called an exaflops – a quintillion of floating point mathematical operations per second. IBM expects to be able to reach this level, but much remains to be done. For the first time, the total performance of all 500 systems on the list exceeded an exaflops – 1.22 exaflops.

But the measurement system used by the Top500, a mathematical problem called Linpack, is an incomplete reflection of all the types of work that a supercomputer could perform. On a wider alternative called High Performance Conjugate Gradients (HPCG), Summit and Sierra took first and second place.

"We have included the HPCG benchmark results in the Top500 list to provide a more balanced overview of performance," organizers said in a June 2018 survey. According to this measure, Summit obtained 2.93 petaflops HPCG and petaflops HP 1.80 HPCG.

China has more supercomputers on the list

Although China lost first place, it expanded its lead as measured by the number of systems on the Top500 list. Compared to the November 2017 list, it went from 201 systems from 500 to 206. The United States went from 143 to 124.

China has grown in recent years to become the first buyer of supercomputers on the Top500 list of the most powerful machines.

Top500

Progress in supercomputers has declined since chip makers like Intel have struggled to increase the performance of general-purpose processors. As a result, many systems are boosted by specialized specialized chips that can speed up specific operations. Summit and Sierra use chips that Nvidia has derived from its GPU business.

The number of supercomputers using accelerator chips has increased from 101 to 110.

Security: Keep up-to-date with the latest violations, hacks, patches, and all the cybersecurity issues that keep you awake.

Blockchain Decoded: CNET is looking at bitcoin technology – and soon, too, a myriad of services that will change your life.

[ad_2]
Source link