AsianScientist (Jan. 30, 2020) – In the highly dynamic fields of high performance computing (HPC) and artificial intelligence (AI),it is easy to get caught up in the frenetic pace of change. But technology is ultimately only a tool. What is needed for HPC and AI to reach their true potential is users and developers who can take the technology forward; in other words, talent.
And yet, it can be daunting for young people hoping to enter the exciting fields of HPC and AI. While many learning resources are now available online, there is still nothing like hands-on experience and real-life datasets. To identify and support the next generation of HPC-AI professionals, the HPC-AI Advisory Council and the National Supercomputing Centre (NSCC) Singapore came together to organize the APAC HPC-AI Competition, now in its second year.
Aimed at university students across the Asia-Pacific, the competition challenges teams to come up with optimized solutions for HPC and AI problems while staying within strict power constraints. This year saw 26 teams participating, ten more than the 16 teams that took part in the inaugural edition the year before. The 2019 competition also saw greater diversity, with new teams from India, Sri Lanka and Malaysia throwing their hats into the ring.
For the HPC component, teams had to run SWIFT, an application for simulating the effects of gravity and hydrodynamics on astrophysical objects such as planets and galaxies. Designed to simulate the entire universe, SWIFT is used by astronomers to test theories of how black holes form and how the universe evolved from the Big Bang up to the present day. Teams were tasked with running SWIFT over 32 nodes and assessed on how much they were able to speed up the performance of the application.
For the AI part of the competition, the students were asked to use machine learning to correctly identify images in the ImageNet dataset and judged on the accuracy they could achieve. Using TensorFlow, an open source software library developed by Google for deep neural networks, the teams were given 90 minutes of training time and up to 32 GPUs to classify the over 1.5 million images in the ImageNet dataset.
“The world is facing a sea change in the way we will operate going forward, and that transformation will definitely be powered by HPC-AI,” said Associate Professor Tan Tin Wee, chief executive of NSCC, addressing the students gathered at the APAC HPC-AI Competition award ceremony held on August 26, 2019.
“Even if you are not the top winner today, all of you are winners already for having made it through so many cycles of competition. You represent the crème de la crème of HPC-AI in the world; whatever you do, your future is extremely bright,” he said.
Reaching a training speed of 11,720 images per second over 16 GPUs and improving the performance of SWIFT nearly three times helped the ‘Bowhivolar’ team from Taiwan’s National Cheng Kung University clinch the first place. We caught up with a member of the winning team, Mr. Liu Yen-fu, to find out what is next for the bright, young students.
What made you join this competition?
Liu Yen-fu: As a first-year student majoring in engineering science, I am interested in programming and saw this competition as an opportunity to improve myself. I was also hoping that it would give me a chance to broaden my experiences overseas.
How do you feel about winning this challenge?
LY: I am very happy that our team performed so well. Of course, we joined hoping to get an award, but we were not expecting to win the first prize as this is an international competition entered by many of the top universities in the world
What is next for you and Team Bowhivolar?
LY: As winners of this competition, we will be given the opportunity to participate in the International HPC-ISC Student Cluster Competition to be held in Germany in June 2020. For this competition, we will have to prepare our own cluster and we don’t have much money for it. Nonetheless, we are excited about the challenge.