News

Articles:

First-ever APAC HPC-AI competition launched at Supercomputing Asia 2018

First-ever APAC HPC-AI competition launched at Supercomputing Asia 2018

Earlier this month, the National Supercomputing Centre (NSCC) Singapore and the HPC AI Advisory Council opened
the 2018
APAC HPC-AI Competition
 for registration. The competition which aims
to foster international exchange in the areas of high-performance computer
(HPC) and artificial intelligence (AI), was launched at the ongoing Supercomputing Asia Conference yesterday.

Established in 2015, NSCC Singapore manages Singapore’s
first national petascale facility with available HPC resources to support
science and engineering computing needs for academic, research and industry
communities. The HPC AI Advisory Council is a leading organisation for
high-performance computing and artificial intelligence research, outreach and
education.

Since HPC and AI applications share the same underlying
technologies and infrastructures, new developments in HPC are immediately being
adopted for AI, and vice versa. Participants in the competition will get the opportunity
to showcase their expertise in these two intersecting fields of HPC and AI.

The competition will continue
through
until August 2018 and it isopen
to university and technical institute teams from the entire APAC region, and
includes both creating missions and addressing challenges around AI development
and testing, and high-performance computing workloads.

The target of this competition is to improve and enhance the
performance of TensorFlow and/or Caffe2 distributed training implementation
over RDMA and GPUDirect technologies.

The competition consists of two parts. The first part
focuses on AI. The mission of Part 1 is to reach the highest distributed
training performance. The teams can choose from two framework options, TensorFlow over RDMA (Remote Direct
Memory Access) and Caffe2 over RDMA.

TensorFlow is
an open source software library for numerical computation using data flow
graphs. Caffe is
a deep learning framework made with expression, speed, and modularity in mind.
It is developed by Berkeley AI Research (BAIR) and by community contributors.

The RDMA implementation code information is available here.
The benchmark is distributed training based on Imagnet dataset

The HPC-AI competition committee has tested the benchmark on
a particular system to set baseline performance criteria. The participating
teams are also asked to run the benchmark to define a baseline on their own
cluster and on the NSCC-available supercomputer. They will be awarded points
based on performance improvement based on the HPC-AI competition committee
results and also on the team’s own baseline.

In the second part,
on HPC,the teams are asked to benchmark the Weather Research and
Forecasting (WRF)
Model, a next-generation mesoscale numerical weather prediction system,
designed to serve both operational forecasting and atmospheric research
needs. 

The competition’s website currently lists 17
participating teams
, from 7 countries (1 from Bangladesh, 5 from China, 1 from Japan, 2 from Korea, 3 from Singapore, 3 from Taiwan and 2 from Thailand).

The winning team will get a reserved spot representing APAC
at the 2019 International
HPC-ISC Student Cluster Competition
, Germany, along with a cash prize of US$4,000.