ARC grant helps build $2 million system for complex modelling and big data

An Australian Research Council (ARC) grant will give students and researchers access to a unique high-end system to help them analyse complex biological systems and big datasets.

The University of Melbourne has used the $690 000 Linkage Infrastructure, Equipment and Facilities (LIEF) grant - along with contributions from partners - to commission Dell EMC to build a scalable, efficient, petascale graphical processing unit (GPU) system with the PowerEdge C4130 compute server, comprising 252 NVIDIA Tesla P100 GPU accelerators and 1,764 Intel Xeon CPU Cores.

This $2 million system will ensure researchers and students continue to have the necessary high-performance computing capacity to analyse and visualise large datasets from scientific instruments needed to produce world-class research in engineering and the biological, chemical and physical sciences.

University of Melbourne Project Lead and  Department of Computer Science and Information Systems Head Justin Zobel said he is keen for students to experience this smart new system.

“We want our graduates to be trained on the latest machines and to be capable of working on complex problems which employ the latest thinking in computing science,” Professor Zobel said.

“Adding to the Parkville precinct’s strategic capability, we are building greater digital competence on a range of new architectures, and developing cutting-edge training programs, which will see our student programmers out working in community engagement projects throughout the precinct and beyond.”

The system will also allow some of Victoria’s top researchers to continue their work in complex molecular modelling for better drug design, computational biophysics for the development of nanomaterials, computational fluid dynamics for industrial and medical device design, and computer science and algorithm development for interrogating complex datasets.

The NVIDIA GPUs will be integrated with the existing 6 PB of storage and 8000 cores of cloud currently available on the University’s shared file systems, ensuring there is both maximum flexibility and major savings, as well as high-performance cloud capacity.

Partners at La Trobe, Deakin and RMIT Universities will also get access to this petascale compute capacity, which has been unavailable to them since the decommissioning of the Victorian Life Sciences Computation Initiative’s IBM Blue Gene/Q at the end of 2016.

Dell EMC high performance computing lead for Australia and New Zealand Andrew Underwood said: “With this new system, the team at Dell EMC are excited for the discoveries and innovation that will be achieved by the University of Melbourne and its partners. Particularly the exciting educational opportunities to come.”

NVIDIA Australia/NZ country manager Mark Patane said: “NVIDIA Tesla GPU accelerators are ideally suited to a wide array of HPC workloads, including simulation, visualisation and deep learning.

"Our collaboration with Dell EMC on this state-of-the-art GPU computing platform will power a new wave of research at the University of Melbourne, and our work with the university will result in training and code acceleration for a broad community of GPU users.”

Existing users will get to trial the system late in 2017, leading up to full production by early 2018.

Technical data

The new PowerEdge C4130 based system will deliver an estimated 834 TFLOPS Rmax and 1.22 PetaFLOPS Rpeak with

63 x PowerEdge C4130 Compute Servers: 4 x NVIDIA Tesla 12GB P100 PCIe GPU; 2 x Intel Xeon E5-2690 V4 CPU; 128GB DDR4 RAM; Mellanox ConnectX-5 100GbE Network; iDRAC8 Enterprise

Total capacity: 1,764 Intel Xeon Cores; 903,168 CUDA Cores; 8,064 TB RAM