ORNL building world’s smartest supercomputer

 

Note: This story was last updated at 10 a.m. Aug. 8.

Oak Ridge National Laboratory on Wednesday said it is building the world’s smartest supercomputer.

The new supercomputer is called Summit. It will be located in a new data center next to Titan, which is now the fourth most powerful supercomputer in the world.

Summit will be 5-10 times faster than Titan, ORNL said. It will move data five to 10 times faster, store eight times more data, and perform many more calculations simultaneously than Titan, the lab said in information provided by spokesperson Morgan McCorkle.

Summit will be the world’s smartest supercomputer because of its enormous memory and data handling capabilities as well as its unique machine learning processor design, McCorkle said.

“The first of Summit’s cabinets arrived Monday, and our team is in the process of uncrating and putting them in place,” McCorkle said in response to questions from Oak Ridge Today.

ORNL has about 600 people involved in high-performance computing, and there was excitement in the supercomputing community as the first new cabinets arrived Monday.

Summit will be in the Oak Ridge Leadership Computing Facility at ORNL. The new data center there includes new infrastructure such as a new mechanical energy plant, as well as substations, transformers, heat exchangers, and cooling towers. ORNL has been preparing for Summit for more than two years through facility upgrades and science application readiness, McCorkle said.

The cabinets that arrived Monday are from ORNL’s vendor partner IBM. Once they are installed in Summit’s room, ORNL will install the internal computational and networking electronic components and connect it to the power and cooling infrastructure, McCorkle said.

Installation of the hardware could take six months or more.

The supercomputer is expected to be fully operational for scientific users by January 2019. Before then, it will be used by teams in ORNL’s early science program, called the Center for Accelerated Application Readiness: https://www.olcf.ornl.gov/caar/.

“Summit will outperform Titan in every way,” McCorkle said. “Researchers who have been using supercomputers like Titan will be able to add much more complexity to their codes, which will produce models and simulations with much greater resolution and higher fidelity for new opportunities to make scientific, energy, and medical discoveries.”

There will be a short overlap period when Summit comes online, and then Titan will be decommissioned to make room for the next supercomputer, which will be Summit’s successor, McCorkle said.

Besides being 5-10 times faster than Titan, Summit will use about 60 percent more floor space. The lab has said that Titan is as big as a basketball court.

Summit is expected to have about 4,600 nodes and still deliver at least five times the computational performance of Titan, which has 18,688 nodes. Oak Ridge Today has previously reported that each Summit node will have central processing units (CPUs), graphics processing units (GPUs), high-speed interconnecting links, and a huge amount of memory.

Some information about Summit—such as total cost, total number of cabinets, and final configuration—hasn’t been released yet.

ORNL has said it has been working with IBM, Nvidia, and Mellanox since 2014 to develop Summit.

An ORNL website page that describes the project says Summit, like Titan, will have a hybrid architecture.

“Each node will contain multiple IBM POWER9 CPUs and NVIDIA Volta GPUs, all connected together with NVIDIA’s high-speed NVLink,” according to the page, part of the Oak Ridge Leadership Computing Facility at ORNL. “Each node will have over half a terabyte of coherent memory (high bandwidth memory plus DDR4) addressable by all CPUs and GPUs plus 800 GB (gigabytes) of non-volatile RAM (random access memory) that can be used as a burst buffer or as extended memory.”

ORNL has previously said that the hybrid system that combined CPUs and GPUs in Titan allowed it to fit into the same space as Jaguar, an earlier supercomputer at ORNL, while using only slightly more electricity. That’s important because supercomputers can consume megawatts of power.

GPUs were first created for computer games.

While Summit will be 5-10 times faster than Titan, Titan is 10 times faster than Jaguar, the computer system it replaced. Jaguar, which was capable of about 2.5 petaflops, had ranked as the world’s fastest computer in November 2009 and June 2010.

Installed in 2012, Titan is capable of 17.6 petaflops, according to the semiannual TOP500 List, which uses a benchmark test known as Linpack. A petaflop is one quadrillion calculations per second. That’s 1,000 trillion calculations per second.

 

Titan is used for research in areas such as materials research, nuclear energy, combustion, and climate science.

Titan, a Cray XK7 system, was the top supercomputer in the world for a short time, in November 2012. But it was bumped to number two in June 2013 and it dropped to number three in June 2016. It was bumped down to number four in June by Piz Daint, a 19.6-petaflop Cray XC50 system installed at the Swiss National Supercomputing Centre.

China continues to have the world’s two most powerful supercomputers: the Sunway TaihuLight, a 93-petaflop system developed by China’s National Research Center of Parallel Computer Engineering and Technology and installed at the National Supercomputing Center in Wuxi, and Tianhe-2 (Milky Way-2), a 33.9-petaflop system developed by China’s National University of Defense Technology and deployed at the National Supercomputer Center in Guangzho.

It’s not clear if there is a difference between being the world’s smartest supercomputer and being the world’s most powerful supercomputer. Oak Ridge Today has sent a few follow-up questions to ORNL, and we will update this story as we learn more.

Oak Ridge Today reported a year ago that Summit is one of three new more powerful supercomputers that the U.S. Department of Energy had planned to start operating soon, possibly as early as this year. Besides Titan at ORNL, the machines that were planned then included Sierra at Lawrence Livermore National Laboratory near San Francisco and Aurora at Argonne National Laboratory in Chicago. The status of Sierra and Aurora wasn’t immediately clear Thursday morning.

More information will be added as it becomes available.

You can learn more about Summit here.

See our previous supercomputer stories here.

See a TOP500 story on Summit here.

 

 

 

 

 

 

Do you appreciate this story or our work in general? If so, please consider a monthly subscription to Oak Ridge Today. See our Subscribe page here. Thank you for reading Oak Ridge Today.

Copyright 2017 Oak Ridge Today. All rights reserved. This material may not be published, broadcast, rewritten, or redistributed.