Home » IT Services News » IT Services News No. 179 (Feb 2016) » Launch of New Heterogeneous HPC Cluster System

Launch of New Heterogeneous HPC Cluster System

We reported in ITS News (No. 176) that a new heterogeneous High Performance Computing (HPC) Linux cluster, HPC2015, has been set up to support both advanced compute- and data-intensive multi-disciplinary research projects of HKU.

The cluster has been put into full production use since October 2015.  All staff and students who require using intensive computational power to support their research projects are welcome to use this new facility.

This new system comprising of 104 general purpose compute nodes are outfitted with the latest Intel Xeon v3 processors which deliver exceptional performance improvements on high performance scientific applications.  The significant performance boost in processing HPC workloads is attributed to Intel AVX2 (Haswell processors), fast I/O (Luster parallel file system), fast interconnect (InfiniBand network) and new utilities and libraries (Compilers, parallel libraries, math libraries, etc).  Software that has already been optimized for accelerated computing can be further speeded up by utilizing the GPGPU/Xeon Phi coprocessors in special purpose compute nodes.

Below is a table comparing the clocked execution time of scientific simulations between Gridpoint System (based on Intel Nahalem 2.53 CPU) and HPC2015 System  (comparison is based on the same number of CPU cores)- 

Software/Application

Execution time in GRIDPOINT

Execution time in HPC2015

Speedup

GROMACS

423.5 hours

18 hours

23.5x

VASP

15 hours

1.25 hours

12x

Siesta

86 seconds

46 seconds

1.9x

The utilization of the system (up to January 2016) is on a steady growth and the usage is illustrated by the following charts-

CPU Usage and User Accounts of HPC2015 System
(Oct 15 – Jan 16)
A Comparison on the Usage of the Current HPC Facilities
(Jul 15 – Jan 16)
CPU Usage and User Accounts Comparison on the Usage

Below is some feedback received from researchers using the HPC2015 System-

“…HPC2015 is much more efficient than gridpoint and hpcpower2.….. For a large job more than 200 atoms, it is impossible to perform on one node on hpcpower2, … but only half an hour on HPC2015. “ – YU Hulei, Mechanical Engineering (Application: VASP)

“HPC2015 system provides sufficient memory for RAM-intensive bioinformatic programs, which could not be handled by the gridpoint system.….. Huge memory node and large disk storage are essential for the de novo assembly and downstream computation…..” – LEUNG Wing Sze, Biological Sciences (Application: Trinity)

“…With the new HPC2015 system, the progress of my project is significantly advanced, which is benefited both from the hardware and the software …… The new cluster also supports AXV2_256 SIMD which also benefits the performance of Gromacs…… Although the number of cores increases 25%, the simulation speed is accelerated by 29 times. I believe that the MD simulation with Gromacs can be further improved by using MIC coprocessor and GPU……” – Dr. ZHANG Chunlei, Chemistry (Application: GROMACS)

The HPC2015 system is particularly suitable for researchers whose projects require extremely high computation requirements (CPU, memory, etc) or for those who would like to explore new solutions with salient hardware features.  For more information about this new system, please visit http://www.its.hku.hk/services/research/hpc/hpc2015  for more details. 

In case you have any questions about the HPC facilities supported by ITS, please contact undersigned or email to GROUP-ITS-HPC@hku.hk.

 

Lilian Chan
High Performance Computing Team
Tel: 3917 2485
Email: lilianyl@hku.hk