Supercomputing will be a critical tool in helping the world avoid climate crisis — but the HPC sector has its own challenges ahead in improving environmental performance, writes Mark Stickells, executive director at Pawsey Supercomputing Centre.
Whether it is helping maximise the output for wave power, investigating new materials for hydrogen, or making legacy gas turbines more efficient, supercomputers are playing a critical role in the energy transition future.
In fact, it’s often argued that the green transition can’t take place without the digital and HPC transition occurring alongside it, driving and refining innovation and testing approaches at speed.
But here’s the rub: many HPC centres are themselves big users of energy, thanks to the requirements of powerful processors and the correspondingly powerful cooling needed to keep them operating.
Running the world’s fastest supercomputer in Japan takes about 29 megawatts of power — 26MW at its most efficient — or enough to power about 40,000 Australian homes.
It’s no surprise, therefore, that the sector is striving to make HPC more energy efficient, speeding up our own energy transition even as we support the science to help other industries do the same.
How big is the challenge?
It is estimated the information technology and computing sector contributes 2% of the world’s carbon dioxide emissions, with data centres and HPC facilities contributing a significant portion of that total.
There is some good news, with estimates that HPC centres are likely to use 620 billion kWh less than was anticipated a decade ago, as both computing and renewable technology improves.
But energy use is still a major challenge, and as more users of supercomputers — major industry, university institutions and governments — have to declare the emissions throughout their supply chain, focus will inevitably turn to how much HPC contributes to that mix.
There’s a philosophical element to this problem, of course.
Because supercomputers exist, and can be put to use in such a breadth of ways, other things do not need to exist.
When a supercomputer like Magnus calculates the best way to maximise the use of gas turbines, that means the researchers behind this science don’t need to build a physical warehouse for that test to take place, for example, or operate machinery to generate those winds, or travel to a remote location where these tests can take place.
In other words, the supercomputer’s energy use is often a real-world energy saver — in this case, a half a million-dollar saving in CO2 emissions — but it is not enough to stop there.
The first step in the greening of HPC is better measurement, tracking and reporting of power use, particularly among the biggest supercomputing centres on the planet.
As we install the first stage of Setonix, which will be the most powerful research supercomputer in the Southern Hemisphere, we have debuted on the Green500 which sits alongside the Top500 supercomputers by power.
Our current supercomputer Magnus rates 453rd in the world for performance (Top500), and Setonix, if fully operational today, would rank in the top 10. And while Setonix delivers 30 times the computational speed of Magnus and Galaxy, we expect it to require only slightly more energy, and our challenge remains to mitigate this use as far as we can.
We will be working with our partners at CSIRO to build a complete picture of the demand and efficiencies Setonix can deliver and will be reporting on its performance to drive improvement over time.
A second step is to look for technological answers to the challenge.
Our centre has been a test case for novel zero emission geothermal cooling, introduced nearly seven years ago. Pawsey uses water from an underground aquifer to cool our supercomputers and recycles the warmer output water through the natural recharge of the aquifer over time.
This is a process that has already saved millions of litres of water since it was introduced, and cut our cooling bill significantly, but we can see from partners around the world that even more revolutionary options are being developed.
In 2021 we signed an MOU with the Finnish team behind LUMI, a pre-exascale supercomputer expected to have a zero or even negative carbon footprint thanks to the use of 100% renewable hydropower, and the transference of heat to warm 20% of a nearby city.
Partnerships like these not only expedite our understanding of how to deliver HPC more effectively, but also how to reduce our environmental impact on the way.
A third measure is to see HPC as part of a broader ecosystem, and look at how the energy demand might provide a network benefit.
There is growing interest in the area of so-called stranded power, or renewable energy that can’t easily be returned to ageing power grids without the risk of overloading them.
Many grids can’t take a sudden influx of power returned by household solar panels, for example, when the sun is out and traditional energy demand is low.
A number of HPC centres are now exploring whether they can provide a home for this stranded power, reducing their baseload demand at the same time.
Hydrogen also offers some exciting possibilities. Woodside has recently announced its plans for a $1 billion blue and green hydrogen plant in Kwinana, which will include converting stranded power into fuel that can be used in other ways. At the same time, we are beginning to see the first hydrogen fuel cells used to provide back-up power for data centres.
There are real possibilities for Pawsey to provide a proof-of-concept implementation and Pawsey is working with providers to investigate the viability and options.
As the world begins to take the necessary steps to reduce fossil fuel dependence, supercomputers will be critical in advancing science and accelerating discovery.
At Pawsey, we are driving the rise of green supercomputing as part of that journey.
About Pawsey Supercomputing Centre
The Pawsey Supercomputing Centre is an unincorporated joint venture between CSIRO, Curtin University, Edith Cowan University, Murdoch University and The University of Western Australia. It is supported by the Western Australian and Federal Governments.
The Centre is one of two, Tier-1, High Performance Computing facilities in Australia, whose primary function is to accelerate scientific research for the benefit of the nation.
Our service and expertise in supercomputing, data, cloud services and visualisation, enables research across a spread of domains including astronomy, life sciences, medicine, energy, resources and artificial intelligence.
We provide support to students and industry personnel, researchers, academics and scientists via seminars and symposia covering data, visualisation and supercomputing, as well as training, internships and workshops covering advanced computing and research and development opportunities.
Source: Mark Stickells, Pawsey Supercomputing Centre