by Heather Mackey

Sometime in the next five years, deep in the Australian outback or in a remote corner of South Africa, work will begin on the largest, most ambitious astronomy project on the planet. The Square Kilometre Array – or SKA – will create a mammoth radio frequency telescope aimed at piercing the mysteries of the cosmos, and NVIDIA GPUs will play a key role in the computational heavy lifting behind it.

SKA’s scope is staggering. At an estimated cost of $2.1 billion, the array will be 10,000 times more powerful than any telescope currently in use and could generate more data in a single day than the entire Internet.

The square kilometer in its name refers to the size of its collecting area, or “dish.” Radio frequency telescopes need far larger collecting areas than optical telescopes, as radio waves are on a much greater spectrum than light waves. But rather than utilizing a one-kilometer-square dish, SKA uses 3,000 small dishes which gather data equivalent to what a single square kilometer dish could accomplish.

When fully operational sometime in 2020, SKA will have the power to capture data on the evolution of galaxies, shed light on dark energy and detect organic molecules in space.

SKA Radio Telescope: By the numbers
Total array surface area 1 square kilometer
Size of each dish 15 meters in diameter
Total number of dishes 3,000
Cost of SKA ~1.5 billion Euro
Estimated launch 2024
Estimated footprint ~6,000 km across

As you might expect, SKA’s computing, networking and storage challenges are extreme. According to one research report from Oxford University, SKA will need an exaflop-capable supercomputer to process data (that is, one that’s 1,000 times more powerful than today’s top performing supercomputers).

Radio interferometers – the type of telescope represented by SKA – correlate signals from two or more antennas. Cross-correlation turns the data into “visibilities,” enabling astronomers to create a picture of the sky.

Previously, the process of correlating signals to form a picture was so computationally intensive that astrophysicists had to rely on specialized hardware. This has raised challenges in scalability and performance within limited time and budgets.

Seeing the limitations of current solutions, Harvard University astrophysicist Michael Clark, along with Paul La Plante and Lincoln Greenhill, explored using GPUs to process radio interferometers’ signal data. Using Fermi-based NVIDIA GeForce GTX480 processors to crunch part of the correlation algorithm called the X-engine, they found that  the X-engine “is a perfect match to the GPU,” according to their published findings.

Clark’s findings have broad implications for radio astronomy. As arrays grow to include huge numbers of antennas, correlation strategies need to evolve. The programmability, flexibility and lower cost make GPUs ideally suited to perform signal correlation.

Already anticipating the needs of SKA, the Australian Commonwealth Scientific and Industrial Research Organisation (known as CSIRO) has begun building out the petascale and real-time supercomputing Pawsey Centre with a GPU cluster for data analysis.

As SKA proceeds, it promises not only to transform how we see our universe, but also to transform supercomputing here on Earth.