Academic supercomputers ‘can be worse for environment than flying’

Efforts to reduce higher education’s carbon footprint should take into account that, for some researchers, universities’ demand for ever greater computing power could be at least as damaging to the climate as flying, academics have warned.

An increasing number of universities have pledged to go zero carbon, but many academics need access to ever faster supercomputers – which demand an increasing amount of energy – in order to stay on the cutting edge of research.

“If you look at my own carbon footprint, which is absolutely horrifying, my flying is bad, but my computing is worse,” said Simon Portegies Zwart, professor of computational astrophysics at Leiden University in the Netherlands.

Supercomputers are generally used in disciplines where researchers need to model extremely complex systems, like galaxies, fluid dynamics and, coincidentally, the earth’s climate.

Their power consumption can be enormous. Japan’s Fugaku supercomputer, the most powerful in the world and used by a host of Japanese universities to explore everything from earthquakes to the evolution of the universe, uses more than 28 megawatts of power, the equivalent of tens of thousands of homes in the United States – effectively a large town.

Last month, in a special issue of Nature Astronomy, Australian astronomers calculated that their supercomputer use accounted for the majority of their carbon emissions – more than triple that of flying.

“It did surprise us,” said Adam Stevens, a postdoctoral researcher at the University of Western Australia, and one of the co-authors.

“We went in with the mindset that flying would be the worst. It wasn’t until we actually had real data...when we realised, wow, this is huge.”

The power consumption of the world’s top 500 computers – a list compiled by academics twice a year – is growing. Fugaku consumes four times as much power as the world’s fastest machine a decade ago, for example.

“I am absolutely pushing in the Netherlands for a bigger supercomputer,” admitted Professor Zwart. Yet while scientists expand the complexity of their simulations to fit the available computing resources, he said, there is “no one asking whether it’s necessary”.

One big caveat is that supercomputers run on national energy grids, meaning their emissions depend on the mix of coal, gas, nuclear and renewables in a country.

This is one of the reasons why astronomers from the Max Planck Institute for Astronomy in Germany, writing in the same journal issue, found that supercomputing accounted for a far lower share of their emissions compared with their Australian colleagues; Germany has a much cleaner energy mix than Australia.

One solution, suggest the astronomers in Germany, might be to build supercomputers somewhere like Iceland: a chilly country, which helps with cooling, that also offers abundant renewable energy.

After all, each country does not boast its own telescopes – astronomers share facilities in countries such as Chile, because of its altitude and clear dry air.

But governments want the supercomputers they fund to be in their territory, said Professor Zwart; there is national competition at play. “It’s political,” he said. Lobbying a Dutch politician to build a more powerful supercomputer in the Netherlands, he recalled being asked: “What are the Belgians doing?”

Universities might not have much influence over national energy policy, but some are taking their power supply into their own hands. In May, Swinburne University of Technology, which houses one of Australia’s major astronomy supercomputers, signed a deal to procure all its electricity from a green power company specialising in wind farms.

Last year, the University of Cambridge also started sourcing a fifth of its power needs from wind farms, said Ian Leslie, chair of the university’s environmental sustainability strategy committee. At Cambridge as a whole supercomputers accounted for just 6 per cent of the university’s electricity last year, he said.

This points to the fact that although supercomputing is a surprisingly dirty business for a subset of researchers, its overall impact is dwarfed by other sources of carbon.

Dan Stanzione, executive director of the Texas Advanced Computing Center at the University of Texas at Austin, argued the issue is a “red herring”.

The world’s supercomputers have an annual carbon footprint in the broad region of 3 million tonnes, he calculated. Of these, roughly half are academic or used by government research labs.

By contrast, beef cattle produced 3 billion tonnes of carbon a year, he said.

“Supercomputers are also just a tiny fraction of all computing,” he added. “Large data centres in the United States alone would emit about 60 million tonnes, so all the supercomputers in the world would be less than 5 per cent of the power we use for email and streaming videos of cats.”

Despite supercomputers’ gargantuan energy consumption, researchers still believe that what they can do is worth it. “Computational modelling is at the heart of understanding and remediating climate change,” said Dr Stanzione. “Without large-scale computational models, we wouldn’t actually know that we need to reduce carbon emissions in the first place.”

david.matthews@timeshighereducation.com