The race to exascale is on — while Canada watches from the sidelines
Scientific discoveries that follow debut of exascale supercomputers may change life in unimaginable ways
This column is an opinion by Kris Rowe, a computational scientist working to get science and engineering applications ready for the next generation of exascale supercomputers. Born and educated in Canada, he has worked at major Canadian and American Universities, as well as two U.S. national laboratories. For more information about CBC's Opinion section, please see the FAQ.
Some of the brightest minds from around the globe have been quietly working on technology that promises to turn the world on its head, but so far Canada has been watching from the sidelines.
While it is unlikely that people will be huddled around their televisions to watch the power to these incredible machines being switched on, the scientific discoveries that follow the debut of exascale computers will change our daily lives in unimaginable ways.
So what exactly is an exascale computer?
It's a supercomputer capable of performing more than a billion billion calculations per second — or 1 exaflops.
"Exa" is the metric system prefix for such grandiose numbers, and "flops" is an abbreviation of "floating-point operations per second."
For comparison, my laptop computer is capable of about 124 gigaflops, or 124 billion calculations per second, which sounds fast.
However, Summit is a mere welterweight relative to an exascale supercomputer, which is more than 60 times faster.
To put that speed in perspective, if you took all the calculations a modern laptop can perform in a single second, and instead did the arithmetic non-stop with pencil and paper at a rate of one calculation per second, it would take roughly 3,932 years to finish.
In a single second, a supercomputer capable of 1 exaflops could do a series of calculations that would take about 31.7 billion years by hand.
While colloquially a supercomputer is referred to as a single entity, it is actually composed of thousands of servers — or compute nodes — connected by a dedicated high-speed network.
You might assume that an exascale supercomputer could be built simply by using 60 times more compute nodes than today's fastest supercomputer; however, the cost, power consumption, and other constraints make this approach nearly impossible.
Fortunately, computer scientists have an ace up their sleeves, known as a GPU accelerator.
Graphics processing units (GPUs) are the professional-grade cousins of the graphics card in your personal computer and are capable of performing arithmetic at a rate of several teraflops (ie. really, really fast). And a feasible route to exascale can be realized by not only making supercomputers larger but also denser.
Sporting six extremely powerful GPUs per compute node, Argonne National Laboratory's Aurora will follow this approach. Scheduled to come online in 2021, Aurora will be the first exascale supercomputer in North America — although the title of first in the world may go to China's Tianhe-3, which is slated to power up sometime in 2020.
Several other machines in the U.S., China, Europe and Japan are scheduled to be brought to life soon after Aurora, using similar architectures
What exactly does one do with all that computing power? Change the world, of course.
Exascale supercomputers will allow researchers to tackle problems which were impossible to simulate using the previous generation of machines, due to the massive amounts of data and calculations involved.
Small modular nuclear reactor (SMR) design, wind farm optimization and cancer drug discovery are just a few of the applications that are priorities of the U.S. Department of Energy (DOE) Exascale Computing Project. The outcomes of this project will have a broad impact and promise to fundamentally change society, both in the U.S. and abroad.
Exascale and Canada
So why isn't Canada building one?
One reason is that exascale supercomputers come with a pretty steep sticker price. The contracts for the American machines are worth more than $500 million US each. On the other side of the Atlantic, the EU signed off on €1 billion for their own exascale supercomputer.
While the U.S. and Europe have much larger populations, the annual per capita spending on large-scale computing projects demonstrates how much Canada is lagging in terms of investment. The U.S. DOE alone will spend close to $1 billion US on its national supercomputing facilities next year, a number which does not take into account spending by other federal organizations, such as the U.S. National Science Foundation.
In comparison, Compute Canada — the national advanced research computing consortium providing supercomputing infrastructure to Canadian researchers — has a budget that is closer to $114 million Cdn.
In its 2018 budget submission, Compute Canada clearly lays out what it will take to bring our country closer to the forefront of supercomputing on the world stage. Included is the need for increasing the annual national spending on advanced research computing infrastructure to an estimated $151 million — a 32 per cent increase from where it is now. Given cost of the American exascale supercomputers, this is likely a conservative estimate.
However, the need for an exascale supercomputer in Canada does not seem to be on the radar of the decision-makers in the federal and provincial governments.
Hanlon's razor would suggest that this is not due to some sinister plot by politicians to punish the nation's computer geeks; rather, our politicians likely don't fully understand the benefits of investing in the technology.
For example, the recent announcement by the premiers of Ontario, Saskatchewan and New Brunswick to collaborate on aggressively developing Canada's small modular reactor (SMR) technology failed to mention the need for advanced computing resources. In contrast, corresponding U.S. DOE projects explicitly state that they will require exascale computing resources to meet their objectives.
Why should the Canadian government — and you — care?
For the less altruistic, a benefit of supercomputing research is "trickle-down electronics." The quiet but persistent legacy of the space race is technology like the microwave oven found in most kitchens. Similarly, the technological advances necessary to achieve exascale computing will also lead to lower-cost and more energy-efficient laptops, improved high-definition computer graphics, and prevalent AI in our connected devices.
But more importantly for Canada, how we invest our federal dollars says a lot about what we value as a nation.
It's a statement about how we value the sciences. Do we want to attract world-class researchers to our universities? Do we want Canada to be a leader in climate research, renewable energy and medical advances?
It's also a statement about how much we value Canadian businesses and innovation.
The user-facility model of the U.S. DOE provides businesses with access to singular resources, which gives American companies a competitive advantage in the world marketplace. Compute Canada has a similar mandate, and given the large number of startup companies and emerging industries in Canada, we leave our economy on an unequal footing without significant investments in advanced computing infrastructure.
Ultimately, supercomputers are apolitical: they can just as easily be used for oil exploration as wind farming. Their benefits can be applied across the economy and throughout society to develop new products and solve problems.
At a time when Canada seems so divided, building an exascale computer is the kind of project we need to bring the country together.
[Note: The opinions expressed are those of the author and do not necessarily represent the official policy or position of Argonne National Laboratory, the U.S. Department of Energy or the U.S. government.]
- This column is part of CBC's Opinion section. For more information about this section, please read our FAQ.