Inside SKA's plan to map the universe using Huawei’s Atlas 900 AI tool


The Square Kilometre Array (SKA) has a celestial ambition. The project aims to discover the origins of the universe through the lens of the world's largest radio telescope.

The telescope will search outer space for signals travelling through the cosmos and could reveal how the first stars and galaxies were formed after the big bang and the nature of the mysterious force of dark energy. It may even answer the question that humans have pondered since the dawn of civilisation: are we alone in the universe?

It will also provide the scientific foundations for more practical applications. Radio astronomers previously created the technology behind WiFi and SKA's discoveries could similarly transform the future of telecommunications, GPS and medical imaging techniques.

These ambitions will be pursued by thousands of satellite dishes and up to a million low-frequency antennas spread across two remote locations in South Africa and Western Australia. These isolated sites were chosen because they experience little interface from cellular phone networks and TV broadcasts, which would pollute the signals that the SKA seeks to detect.

Their efforts to gather as much information as possible from the universe will generate vast volumes of data, which will be processed by a high-performance supercomputer in each of the two sites. These machines will have a peak speed of 100 petaflops, which would make them among the fastest supercomputers in the world.

"This is a data-intensive process that will initially require around 50 petaflops of dedicated digital signal processing power, growing to 250 petaflops as our capability increases," Philip Diamond, the director-general of SKA and a professor in the school of physics and astronomy at the University of Manchester, explained at the Huawei Connect conference in Shanghai last week.

"In total, every year we will archive around 600 petabytes of data, more than the amount currently handled by platforms such as Facebook and Google."

Read next: Exscientia CEO explains how AI is transforming drug discovery

Huawei is helping to process all this data through its new Atlas 900 AI training cluster. The system combines the power of thousands of Ascend processors to create what Huawei deputy chairman Ken Hu claims is "the world's fastest AI training cluster."

“Using the ResNet-50 model, which is industry standard for measuring AI training performance, we put our Atlas 900 to the test," he said."It finished the entire training in less than a minute – 59.8 seconds, to be precise. This is about 10 seconds faster than the previous world record."

Huawei applied the Atlas 900 cluster to a sky map of the Southern Hemisphere containing 200,000 stars, which was created with data from SKA radio telescopes. Finding a celestial body within an area this big would typically take an astronomer 169 working days. Atlas was able to locate and identify a specific type of star in roughly 10 seconds.

Read next: ALD pioneer Tuomo Suntola explains how his invention sustains Moore's Law

The Atlas 900 could play a key role in making the data generated by SKA's radio telescopes accessible. Prototype instruments have already been deployed on site and official construction is scheduled to start in 2012, with science verification results expected in the mid-2020s.

"The excitement and momentum within the SKA partnership is is building as we approach the start of construction," said Diamond. "And I look forward to the day in 2025 when scientists from around the world, including here in China, will get their first glimpse of the universe through the eyes of the SKA."

This story, "Inside SKA's plan to map the universe using Huawei’s Atlas 900 AI tool" was originally published by

Copyright © 2019 IDG Communications, Inc.

Shop Tech Products at Amazon