The power of exascale computing is measured, unsurprisingly, in ones and zeros, specifically, 1,000,000,000,000,000,000, or 1018. That’s the number of floating-point operations that exascale supercomputers can make in one second, which represents a massive leap forward in computing capability.
According to Hewlett Packard Enterprise (HPE), however, “exascale is more than a speed milestone or a system size. It’s new workloads brought on by new questions intersecting with new compute capabilities to create a major technological shift.”
What Can Exascale Do?
With the capacity to perform quintillion calculations per second, exascale supercomputers will “more realistically simulate the processes involved in precision medicine, regional climate, additive manufacturing, the conversion of plants to biofuels, the relationship between energy and water use, the unseen physics in materials discovery and design, the fundamental forces of the universe, and much more,” says the Exascale Computing Project (ECP).
Supercomputing, HPE says, already plays a key role in the search for a COVID-19 vaccine as well as in extreme weather prediction. “Using supercomputers, for example, forecasters can do a far better job of predicting where a hurricane will hit land―from within 200 miles 50 years ago to just 50 miles today. With exascale speed, future models will produce even greater accuracy and do so faster, saving lives and limiting economic losses as the world confronts the effects of climate change,” says Dr. Eng Lim Goh, senior vice president and CTO for AI at HPE.
Exascale Shift
The U.S. Department of Energy (DOE) plans to roll out three new exascale systems over the next three years:
- Aurora at Argonne National Laboratory and Argonne Leadership Computing Facility (ALCF)
- Frontier at Oak Ridge National Laboratory (ORNL)
- El Capitan at Lawrence Livermore National Laboratory (LLNL)
LLNL’s Jeremy Thomas describes the power of exascale this way:
Exascale is a full 10x order of magnitude more powerful than today’s fastest supercomputers. To put this in perspective, a supercomputer running at one exaflop [or one quintillion computing operations per second] is about a million times more powerful than your typical laptop. At more than 2 exaflops, LLNL’s El Capitan will be faster than the top 100 supercomputers that currently exist in the world—combined.
Modern supercomputing, according to HPE, “relies heavily on technology that allows the millions of the processor cores involved in a single system to be interconnected and effectively communicate.” Rick Stevens, leader of Argonne's Exascale Computing Initiative, explains how the shift to exascale-level capability came about:
One of the things that's helped to drive exascale is the shift from CPUs to GPUs, which has been accompanied by a dramatic improvement in high-bandwidth memory. If we'd tried to power the level of computing GPUs provide with normal memory, it wouldn't work. Exascale systems have a much higher ratio of compute capability relative to memory and bandwidth.
Challenges and Applications
Stevens describes three major challenges that had to be overcome on the road to exascale, including:
- Power consumption: With the components available 10 years ago, an exascale system would have required a gigawatt of power costing US$1 billion per year, meaning more energy-efficient electronics had to be developed.
- Scale: The system uses close to a hundred thousand GPUs, each of which has tens of billions of transistors cycling a billion times a second. Until recently, he says, getting to that many computational elements was impossible.
- Reliability: A system involving hundreds of thousands of electronic components means redundancy and fault tolerance must be built in to keep it working when parts fail.
At the same time, Stevens notes, planning for exascale meant addressing the gap between building the hardware and having the software available to run on it. He likens this effort to “building an airplane while you're trying to fly it, while at the same time you're also building the aircraft carrier where it's going to land.” To address the issue, the DOE created the ECP in 2016 to build exascale-capable applications.
The ECP is developing an expanded and vertically integrated software stack “to include advanced mathematical libraries and frameworks, extreme-scale programming environments and tools, and visualization libraries,” according to an article at The Royal Society. This investment in software libraries and tools, Stevens says, will make it much easier to build new exascale applications.
These applications, The Royal Society article states, encompass broad-ranging areas including:
- Chemistry and materials
- Energy production and transmission
- Earth and space science
- Data analytics and optimization
- National security
Backed by the power of exascale, such applications will lead to major advancements in understanding and have a profound impact on our future.
Comments