CONTENTS
New Projects
Fall 2012
PREVIOUS NEXT
High Performance Computing at HKUST
by Charles Choy
The New HPC Cluster

With increasing demand for high performance computational power, the University has formed a High Performance Computing (HPC) Working Group in 2010 to explore how high performance computing needs in HKUST can be met. The Working Group concluds that needs could be met by two means: (1) setting up an in-house HPC facility at HKUST and (2) renting computational power from external supercomputer centers:

  1. In-House HPC Facility

    With funding support contributed by the Vice President for Research and Graduate Studies, School of Science, School of Engineering and others, a CPU cluster and a GPU cluster have been acquired. The details are as follows:

    • CPU cluster
      • 27 x high-end compute nodes each equipped with
        • Dual Intel Xeon E5-2670 CPU
        • 64G RAM
        • 40G Infini-Band network connection
      • 20 x standard compute nodes each equipped with
        • Dual Intel Xeon E5-2670 CPU
        • 32G RAM
        • Gigabit Ethernet network connection
      • 96 TB disk storage

    The total number of compute cores is 752 with theoretical peak performance of 15.6 TFLOPS

    • GPU cluster
      • 11 x GPU compute nodes each equipped with
        • Dual Intel Xeon E5-2650 CPU
        • 64G RAM
        • 4 x NVIDIA Tesla M2090 6GB module

    The total number of CPU and GPU cores are 176 and 22,528 respectively. The theoretical peak performance is 29.26 TFLOPS

    The CPU cluster have been delivered in Aug and the GPU cluster is expected to arrive in this Oct. They will be installed and managed by ITSC. The arrangement is a continuation of the community cluster project which ITSC initiated in 2009.

  2. Rental of Computational Power

Although purchasing our own on-site HPC resources allows convenience and ease of access, this strategy is hard to scale up and less flexible to cope with sudden surge in demand. Renting computational power from outside becomes necessary when demand is large or surging.

In early 2012, ITSC formed a partnership with the National Supercomputing Center at Shenzhen (NSCCSZ) to provide HKUST users with a free trial period of 6 months. More than 30 accounts were created. In May 2012, staff from NSCCSZ came to HKUST campus and delivered a workshop in HKUST to introduce their services. In return, a team of HKUST delegates visited NSCCSZ in Jun. Although the computing environment still has stability and reliability problems, many of our users manage to leverage this resource. The trial arrangement has already ended and the University is currently working with NSCCSZ for a longer term arrangement.

In the mean time, the University will continue to explore similar collaboration opportunitis with external supercomputing centers if deemed appropriate.

Please do not hesitate to contact us if you have further queries.

Engineer Working on the New HPC Cluster
Delegates from NSCCSZ
Visit to NSCCSZ