The Summit supercomputer will have roughly double the power of China’s top system, according to the DOE
The US Department of Energy (DOE) is planning to bring a supercomputer online in 2018 that will have double the computing power of China’s top system, returning the US to the top of the global charts, the agency has said.
The system, called Summit, was announced in 2014 when the DOE awarded $325 million (£247m) to Oak Ridge National Laboratory in Tennessee to construct it, with part of the funding also going to Lawrence Livermore National Laboratory in California to build a less powerful system called Sierra.
Sierra is, like Summit, due to be completed in 2018, while a third system called Aurora is being built at the Argonne National Laboratory in Illinois and is due to come online by the end of 2018.
China’s Sunway TaihuLight debuted at the head of the TOP500 list of the world’s most powerful supercomputers last week, meaning the country now has the top two systems on the list.
China’s Tianhe-2, an Intel-based system that has held the top spot on the past six bi-annual rankings, is now No. 2 on the list.
Sunway TaihuLight reached 93 petaflops on the Linpack benchmark used by the rankings and has a theoretical maximum speed of 124.5 petaflops, while Summit is set to be capable of 200 petaflops, the DOE said.
“Upon completion, Summit will allow researchers in all fields of science unprecedented access to solving some of the world’s most pressing challenges,” Oak Ridge Laboratores said in a statement on Summit’s website.
Five times Titan’s power
The system is to deliver about five times the performance of the lab’s current supercomputer, called Titan, using only about 3,400 nodes compared to Titan’s 18,688 nodes, the lab said.
Each node is built of multiple IBM Power9 CPUs and Nvidia Volta GPUs, connected with Nvidia’s NVLink, according to the lab.
Sierra is planned to reach 150 petaflops using a similar architecture to Summit, while Aurora is to use Intel Knights Hill chips to reach an expected 180 petaflops, according to the DOE.
China’s system uses its own chips, called ChenWei, developed by Jiangnan Computing Research Lab in Wuxi, with each processor containing 260 cores capable of over 3 teraflops. The system uses a total of 10.65 million cores.
China’s investment into high-performance computing was spurred by the US Commerce Department’s ban last year on the sale of Intel chips to some parts of China, and the result is that the country’s supercomputer development has accelerated much faster than that in the US, according to TOP500 editor Jack Dongarra.
“China has continued in leapfrogging the U.S. by a considerable amount,” he said in a report on the new system.