How do you get the latest news from the heart of a dying star?
You can travel there in reality, but the journey will be lengthy and rather warm at the end. Or you can travel there virtually – thanks to the extraordinary powers of Summit, the fastest supercomputer on earth.
Summit, and its sister supercomputer Sierra, are the result of a $325 million US Department of Energy contract awarded in November 2014 to IBM, Nvidia, and Mellanox. Summit is tasked with civilian scientific research and is located at the Oak Ridge National Laboratory (ORNL) in Tennessee. Summit, a system eight times more powerful than ORNL’s previous system is providing scientists with incredible computing power to solve challenges in energy, artificial intelligence, human health, and other research areas that were simply out of reach until now. Sierra is designed for nuclear weapons simulations and is located at Lawrence Livermore in California.
Since November 2018, Summit has been the fastest supercomputer in the world, capable of 200 petaflops. It’s the first supercomputer to reach exaflop speed (a billion billion operations per second) – achieving 1.88 exaflops during a recent genomic analysis – and is expected to soon reach 3.3 exaflops using mixed precision calculations instead of the double-precision calculations used in the traditional LINPACK benchmark.2 In June 2019, Summit was still the fastest supercomputer in the world, capable of 200 petaflops. It’s also the third most energy efficient supercomputing system, at 14.719 GFlops per watt. 3 And in the just announced Top 500 rankings, Summit and Sierra continue to be the two fastest supercomputers in the world. 4
[IBM also recently announced the most powerful Commercial supercomputer]
So what have Summit and Sierra been doing since IBM Systems Lab Services completed the installation, which required roughly 40 technical consultants and 20,000 hours of work?
Sierra is being used by the National Nuclear Security Administration for predictive applications in stockpile stewardship, helping to assure the safety, reliability, and effectiveness of US nuclear weapons.
Summit, on the other hand, is a civilian resource and gets to tackle some of the biggest problems in science. Researchers are currently using Summit for modeling and simulation in fields such as medicine, climatology, and yes, exploring the hearts of dying stars.
For example, Bronson Messer and his team have been using supercomputers at Oak Ridge for over 15 years to perform computational stellar astrophysicists. They’re interested in what elements stars produce as they live and die. When asked what Summit may help the team solve that they have not been able solve previously, he explains:
“We’d like to more accurately model nuclear burning that takes place in a supernova, which requires multiple scales of physics calculations, from large-scale hydrodynamics (fluid motion) to small-scale particle interactions. Summit is really going to let us crank up the realism on those small scales. We expect to increase the number of elements simulated in nuclear burning from about 13, which is the state-of-the-art today, to about 160.” 5
Exploring supernovae isn’t the only research currently leveraging the power of the world’s fastest supercomputer. Gina Tourassi and her team at the ORNL Health Data Sciences Institute are combining supercomputing and deep learning to advance cancer treatment planning. They leverage AI to enable automated and accurate capture of important cancer surveillance data elements from clinical text documents, then use this information to essentially train computers to read medical documents and extract important information to help doctors determine the best cancer treatments.
And CS Chang is developing complex code to model the behavior of plasma – the hot gas medium in a fusion reaction. Understanding plasma behavior at a fundamental level is critical for research that may one day help meet global energy demand by using seawater to fuel plasma in a fusion device without contributing to the greenhouse effect or producing long-term waste.
Supercomputers Optimized for AI
Summit and Sierra have been described as optimized for artificial intelligence (AI), but what does this really mean? In the case of Summit and Sierra, the answer is in the architecture. Summit uses an advanced “data centric” approach that minimizes data in motion. Its IBM AC922 systems link more than 27,000 NVIDIA Volta GPUs with more than 9,000 IBM Power9 CPUs to provide unprecedented opportunities for the integration of AI and scientific discovery.
Training AI algorithms requires processors that can handle a mathematical workout. Each of Summit’s 4,608 nodes contains six deep-learning–optimized GPUs packed with more than 21 billion transistors. And because deep learning requires less precision than traditional scientific computing, Summit holds the potential to deliver exascale-level performance for AI algorithms that scale.
[Also learn how intelligent simulation exploits AI to improve HPC results.]
Like real estate, the value of a supercomputer’s memory is closely tied to location. Summit’s sizable local memory, including high-bandwidth memory on each GPU, gives AI researchers a convenient launching point for data-intensive tasks. Minimal data movement means researchers can run deep-learning networks faster and achieve greater accuracy.
Clearly, the news from America’s fastest supercomputing facilities is fascinating and important. These discoveries will help shape our understanding of the universe, bolster US economic competitiveness, and contribute to a better future.
 Wikipedia: Summit (supercomputer) https://en.wikipedia.org/wiki/Summit_(supercomputer)
2, 4 Top500: https://www.top500.org/lists/2019/06/
3 Green500: https://www.top500.org/green500/lists/2019/06/
5 Oak Ridge National Laboratory Leadership Computing Facility: Summit https://www.olcf.ornl.gov/summit/