NTU Distinguished themselves in ISC'18

The NTU team consists of 6 undergraduate students and an advisor. 2^nd from left: Zhang Xinye, Tan Ying Hao (Vice-Captain), Heng Weiliang, Shi Ziji (Captain), Wu Bian, Hao Meiru.

The School of Computer Science and Engineering students hit the mark again at the recent Student Cluster Competition held in conjunction with International Supercomputing Conference 2018 (ISC’18) at Frankfurt, Germany from 24 -28 June. 

The team from School of Computer Science and Engineering, distinguished themselves by taking the 2^nd overall score at the Student Cluster Competition held in conjunction with ISC’18, Frankfurt. The competition was intense with some major universities with long tradition in HPC.  In total there were 12 teams from 9 countries such as South Africa, China, Poland, UK, etc. 

The NTU team consists of 6 undergraduate students and an advisor: Shi Ziji (Captain), Tan Ying Hao (Vice-Captain), Wu Bian, Zhang Xinye, Heng Weiliang, Hao Meiru, and the advisor is Assoc Prof Bu-Sung Lee, Francis. During the competition, the teams are required to optimize and run a number of applications from different domains, such as fluid dynamics, quantum physics, and computer vision. 

Shi Ziji, captain of the team, said that it was a hard-won award. They met regularly in weekly discussion and tuned the application and system. To cater to various application requirements, the team took a different hardware configuration, favouring a heterogeneous cluster unlike majority of the other teams which have a homogenous cluster. The NTU configuration has the least number of CPUs and GPUs within the top-scoring teams, yet it worked out well in the competition.

Tan Ying Hao, vice-captain of the team who oversaw the Deep Learning application, commented that it was very challenging to tune the TensorFlow software to maximize the speed of learning on the ImageNet dataset using the NVIDIA GPU cards while ensuring that the cluster stays within the power limit of 3kw, a constraint set by the organizers. With only 10 GPU cards, Team NTU achieved stunning 3170 images per second on VGG16 model, beating many teams who brought 16 GPU cards.

NTU team participation in the competition was only possible with the sponsorship from DellEMC, Nvidia Technological Center Singapore, National Supercomputing Center Singapore, Mellanox Technology, and JOS. Also thanks to SCSE for the support, especially Assoc Prof Francis Lee and Ms Irene Goh from Parallel and Distributed Computing Lab. Also, thank to Asst Prof Lin Guosheng for his guidance on Deep learning challenge.