Author Topic: The Race for the Fastest Supercomputer  (Read 228 times)

Ayoshi

  • Timawan
  • Boffin
  • *
  • Posts: 3055
    • View Profile
The Race for the Fastest Supercomputer
« on: July 01, 2018, 07:35:10 AM »
From: GCN - Nov 29, 2017
Quote
The United States’ supercomputing standing has eroded, but that could change soon.

The top slot in the latest list of the 500 fastest supercomputers released this month is held by the Sunway TaihuLight in China’s National Supercomputing Center in Wuxi. The Sunway TaihuLight has held the No. 1 spot in the last four semi-annual rankings and is significantly more powerful than the other systems on the list, according to Jack Dongarra, a professor at University of Tennessee, Knoxville, who has an appointment at Oak Ridge National Lab and helped created the list.

“If you add up the [petaflops of the] next five machines -- numbers two through six -- then it’s equal to the No. 1 machine,” Dongarra said.

Sunway TaihuLight, which has more than 10 million cores and over a million GB in memory, reached 93,014.6 TFlop/s on the Linpack performance test. It was built by the National Research Center of Parallel Computer Engineering & Technology and uses Chinese-developed ShenWei processors rather than the Intel, IBM or NVIDIA processors used in most other supercomputers.

< snipped >

China’s dominance on the recent lists indicates it is investing heavily in high-performance computing as supercomputers become a vital tool for simulating everything from nuclear explosions to medicine.
« Last Edit: July 01, 2018, 07:54:02 AM by Ayoshi »

Ayoshi

  • Timawan
  • Boffin
  • *
  • Posts: 3055
    • View Profile
Re: The Race for the Fastest Supercomputer
« Reply #1 on: July 01, 2018, 07:51:58 AM »
The U.S. once again has the world’s fastest supercomputer. Keep up the hustle. | The Washington Post -
June 25
Quote
The United States has knocked China out of the No. 1 position in supercomputing. This week, when the latest ranking of the 500 fastest supercomputers in the world was released, the Energy Department’s new Summit machine reclaimed a distinction that China has held for five years. The development is more than a matter of national pride; supercomputers are an indispensable tool for national security, technological progress and economic competitiveness.

How fast is the Summit? To begin with, it is roughly eight times faster than the previous U.S. titleholder, the Titan, from 2012. The Summit, developed for the Oak Ridge National Laboratory in Tennessee (where I work), has a peak performance capability of 200,000 trillion “floating point operations” — or petaflops — per second. That won’t mean much to non-computer scientists, so think of it this way: The entire population of Earth would have to compute continuously for 305 days, performing one operation per second, to match what the Summit does in one second. The Summit exceeds China’s fastest supercomputer by about 30 percent, prompting its ranking by TOP500, a project that I have been involved with since its inception in 1993, along with my colleagues Erich Strohmaier and Horst Simon of Lawrence Berkeley National Laboratory and Martin Meuer of Prometeus, a German technology company.

Supercomputers are systems that harness the power of multiple refrigerator-size units — the Summit uses an IBM system composed of 256 such cabinets, weighing a combined 340 tons and occupying 5,600 square feet — or about the size of two tennis courts. The development of supercomputers was fueled in the 1990s by the Energy Department’s desire to maintain the readiness of America’s nuclear stockpile without actual detonation testing. That required computer simulations capable of modeling nuclear processes down to tiny fractions of a second. No computer on the planet was capable of such precision, so the department embarked on a campaign that would raise the processing speed of the world’s best computers by a factor of 10,000.

=====

From: techcrunch.com

Quote
Summit, which has been in the works for a few years now, features 4,608 compute servers with two 22-core IBM Power9 chips and six Nvidia Tesla V100 GPUs each. In total, the system also features over 10 petabytes of memory. Given the presence of the Nvidia GPUs, it’s no surprise that the system is meant to be used for machine learning and deep learning applications, as well as the usual high performance computing workloads for research in energy and advanced materials that you would expect to happen at Oak Ridge.

IBM was the general contractor for Summit and the company collaborated with Nvidia, RedHat and InfiniBand networking specialists Mellanox on delivering the new machine.