Wednesday, 1 February 2017

New supercomputer transforming research

It is region's most powerful machine and can boost work in various disciplines
By Lin Yangchen, The Straits Times, 31 Jan 2017

A new supercomputer in Singapore, much more powerful than anything else in the region, is transforming research and industry here.

Not yet a year old, the machine is already being used to design safer and more cost-effective megastructures, as well as to understand how human genes cause disease. And it has the potential to transform research in many other areas.

Named ASPIRE 1, it is the heart of the National Supercomputing Centre (NSCC) Singapore jointly set up last year by the Agency for Science, Technology and Research, Nanyang Technological University (NTU), National University of Singapore (NUS) and Singapore University of Technology and Design.

Perched on the 17th floor of one of the Fusionopolis towers in Buona Vista, the machine made by Fujitsu has more than 30,000 cores, compared with two or four in a personal computer. It can do a thousand trillion floating point operations - decimal number calculations - per second.

In November last year, ASPIRE 1 was ranked No. 115 in the world by TOP500, a widely recognised ranking of supercomputers.

In comparison, the world's fastest supercomputer, China's Sunway TaihuLight, has more than 10 million cores and purrs at about 100 times the speed of ASPIRE 1.

But in South-east Asia, nothing comes close to Singapore's latest.

It has even inspired a new biannual magazine supplement, Supercomputing Asia, which made its debut here earlier this month, launched by parent magazine Asian Scientist.

One of ASPIRE 1's early adopters is Professor Allan Ross Magee's group at the NUS Faculty of Engineering, which is using it to help design floating oil storage tanks in a project supported by JTC Corporation.

The supercomputer allows them to simulate the complex movements of multiple oil storage tanks - which are affected by both the water currents around them and the oil sloshing around inside them - more accurately in space and time, allowing them to make the design safer and cheaper.

Prof Magee said: "It allows us to see finer-scale structures which influence the physics in ways we didn't expect... the design, the way it's built and the way it's engineered are affected by computational power.

"A smaller computer would give a less optimal solution within the time limits of the project. We then have to 'over-design' the system, which increases cost."

Another early user of ASPIRE 1 is NTU's Professor Stephan Schuster, who leads the international GenomeAsia 100K project to analyse the complete DNA sequences, or genomes, of 100,000 Asians.

The non-profit consortium hopes to use the information to boost medical advances for the Asian population.

His team has an enormous task for the computer - to swallow 100 billion statistical data points on the genome of each person and compare them with a database.

Shortly after ASPIRE 1 was set up early last year, the researchers ran it at full throttle with genetic data from 2,000 people at the same time.

"We could give them (NSCC) a computational problem of a very big size. We were a stress test for their infrastructure," said Prof Schuster. "Once you put a load on the system, you can see how robust it is. Does power management work, does heat management work, is the speed of data transfer what it should be?"

Although there were teething problems with data storage, the computer did its job.

In fact, the 2,000 genomes used to test the computer have already yielded new discoveries.

The data brought to light the genetic diversity of Asia's hunter-gatherer populations, which Prof Schuster said are worth preserving not only for cultural heritage but also to potentially help overcome future challenges like infectious disease outbreaks.

Powerful as it is, ASPIRE 1 is really enough for the needs of only a single university the size of NUS or NTU, noted Prof Schuster. As soon as the computer became available last year, requests to use it went "through the roof" from both academia and industry, he added.

He estimates that the capacity of the supercomputer should be increased 10 times over the next five years. To him, there is no doubt that this is the way to go. To use commercial computing services for his genome project would cost several million dollars.

"Without the NSCC, we could not do what we're doing now at all... there would be no project."


No comments:

Post a Comment