October 26, 2017
Written by: Allan Williams, Associate Director, Services and Technology, National Computational Infrastructure.
Supercomputers are the giants of the computer world. Every second, they perform trillions of calculations. They fill vast spaces, requiring advanced cooling systems to keep them running under 95 degrees Celsius, 24 hours a day. But this is more than a technology story. Supercomputers power the scientists of Australia, helping them discover new frontiers.
Raijin, is hosted at National Computing Infrastructure
Australia’s fastest supercomputer, Raijin, is hosted at National Computational Infrastructure, in Canberra. Raijin is the name-sake of the Shinto God of thunder and lightning, which seems very apt as it storm-fires at 85,000 cores, 24 hours a day. At its peak, Raijin’s performance is equivalent to 40,000 desktop computers running simultaneously. While it is the most powerful computer in the Southern Hemisphere, we still have a lot of work to do to match the world’s leader, China’s TaihuLight, which harnesses 10 billion cores.
Some of the cool things that supercomputers across the world are working on solving include:
- Unlocking the ‘dark DNA’ of genomes to develop personalised cancer treatments
- Simulating nature
- Shedding light on dark energy
- Developing new climate models
- Tracking Australia from space
- Predicting the more than 1,000 thunderstorms that occur on the Earth’s surface at any moment
Exploring these problems is incredibly complex, and can only be achieved with supercomputers which help researchers simulate and explore the billions of molecular interactions that happen in the world around us. Researchers build these simulations and constantly make small refinements to the models – a slow and labour-intensive process – until they deliver results, and ultimately drive scientific breakthroughs.
Scale and power
To put into context the sheer scale and power required to build these simulations – The number of molecules in just ten drops of water is roughly the same as the number of stars in the universe. Yet to model one single star explosion – which offers incredible insights into things like physics, gravity and weather – would take more than three years just to download the data on a normal desktop computer.
In the case of health, researchers simulate things like what happens when a cancer protein attaches itself to a new site in the body, so they can attempt to intervene in the interaction with a potential new targeted drug or treatment method. Another example is quickly identifying the genetic variant that is causing a child’s illness so they can start appropriate treatment as soon as possible. Pinpointing the genetic variant in the 120 billion nucleotide sequences in our genetic makeup can now be achieved in less than a day on a supercomputer, instead of weeks or months.
NCI National Computational Infrastructure, ANU. Canberra.
For scientific breakthroughs, continued efficiency gains are crucial. When you think of speed you can think of size. The greater the efficiency, the greater number of simulations a scientist can explore, fundamentally accelerating the rate of scientific discovery.
Whilst current systems are impressive, we must keep advancing our supercomputers. The latest introduction of IBM Power Systems into Raijin has helped us edge efficiency forward by injecting gains in some applications of up to three times the performance as compared with the latest x86 architecture. But not all applications are made equal. And that’s another point why Raijin is unique – it is a fully heterogeneous environment for Australian researchers. Meaning it’s a one-stop shop for researchers to use the system that’s going to be the most powerful for their application. By providing the widest possible variety of high-performance computing for our users, researchers will always have the best tools available for their simulations and computations.
Supercomputers really are helping solve some of the world’s greatest societal challenges. And we are on the verge of breaking new barriers. The next frontier? To simulate the human brain at a neural level – which includes the interactions of upwards of 1.73 billion nerve cells and more than 10 trillion synapses, and that’s just a fraction of what we know about the brain today. To simulate the human brain, we’d need a supercomputer to perform at a billion billion calculations per second. Now that would be super.