After five years playing second fiddle to China, the US once again has bragging rights for the world’s fastest supercomputer with a gargantuan IBM machine called Summit at Oak Ridge National Laboratory in Tennessee.
Summit’s performance was revealed Monday on the Top500 supercomputer list, which a group of academic researchers updates twice each year, with a speed score of 122.3 quintillion mathematical operations per second, or 122 petaflops. To match that speed, each person on the planet would have to perform 16 million calculations per second.
The Summit system, with 9,216 IBM processors boosted by 27,648 Nvidia graphics chips, takes as much room as two tennis courts and as much power as a small town. It’ll be used for civilian research into subjects like materials, cancer, fusion energy, fusion energy, astrophysics and the Earth’s changing climate.
Another system called Sierra, somewhat smaller but built of the same components, will be used for nuclear weapons research at Lawrence Livermore National Laboratory in California. It claimed third place with a speed of 71.6 petaflops, trailing China’s Sunway TaihuLight with a 93-petaflop speed.
Supercomputers are high-prestige but high-cost systems. The US Energy Department’s Coral program to build Summit and Sierra, for example, cost $325 million. But they’ve played an important role in everything from modeling nuclear weapons without actual explosions and simulating the universe with a method called computational cosmology.
Building a better benchmark for supercomputing
There’s a national race to a performance threshold called one exaflops — a quintillion floating-point mathematical operations per second. IBM thinks it can reach that level, but there’s a ways to go yet. For the first time, the total performance of all the 500 systems on the list surpassed one exaflops — 1.22 exaflops.
But the measurement system the Top500 uses, a math problem called Linpack, is an incomplete reflection of all the types of work a supercomputer might tackle. On an broader alternative called High Performance Conjugate Gradients (HPCG), Summit and Sierra took the first and second places.
“We have incorporated the HPCG benchmark results into the Top500 list to provide a more balanced look at performance,” the list organizers said of the June 2018 list. By that measure, Summit scored 2.93 HPCG petaflops and Sierra 1.80 HPCG petaflops.
China has more supercomputers on the list
Although China lost the top spot, it widened its lead as measured by the number of systems on the Top500 list. Compared with the November 2017 list, it advanced from 201 systems of the 500 to 206. The US dropped from 143 to 124.
Progress in supercomputers has tailed off as chip manufacturers like Intel have struggled to increase the performance general-purpose processors. As a result, many systems are boosted by specialized chips suited that can accelerate specific operations. Summit and Sierra use chips that Nvidia spun off from its graphics processor business.
The number of supercomputers using accelerator chips increased from 101 to 110.
Security: Stay up-to-date on the latest in breaches, hacks, fixes and all those cybersecurity issues that keep you up at night.
Blockchain Decoded: CNET looks at the tech powering bitcoin — and soon, too, a myriad services that will change your life.