The University of Queensland has purchased a new supercomputer that is faster, multifaceted and more efficient than its current high-performance computers (HPCs).
The HPC, named ‘Bunya’, after the native South-East Queensland tree, was bought from Dell Technologies and is estimated to become operational in July this year.
Professor David Abramson, Director of UQ’s Research Computing Centre (RCC), said Bunya can perform across a wide range of research domains, from the sciences to the humanities.
“Bunya will strengthen UQ’s position as a tier-2 supercomputing capability,” Professor Abramson said.
Jake Carroll, RCC Chief Technology Officer, said Bunya will help maintain UQ’s competitive advantage.
“Applications that are communication intensive, or move lots of data as part of their workflow, could run many times faster on Bunya, when compared to our previous platforms, significantly shortening the time to reach research discoveries,” Mr Carroll said.
“As scientific instruments output more, as models grow and as in-silico exploratory techniques expand, so must the resources to support them.
“Bunya is more power efficient than our current HPCs and will allow us to minimise our environmental impact.”
Bunya will replace three of UQ’s older HPCs, Awoonga, FlashLite and Tinaroo, having served the University for almost seven years.
Awoonga was decommissioned in March, FlashLite will be shut down mid-year, and Tinaroo later in 2022.
Wiener, UQ’s four-year-old imaging-intensive, GPU-enhanced supercomputer, will continue to operate.
UQ has allocated funding to expand Bunya’s capabilities and capacity over the next few years, so the single system should be a sufficient replacement for the three HPCs.
“RCC will be able to provide a better, more efficient experience long term in this new consolidated model with one HPC. As time has passed, the need and validity of differentiated supercomputers has waned,” Mr Carroll said.
Mr Carroll said it is possible Bunya’s users may see substantially higher performance than previous UQ RCC supercomputers, depending upon the software users are running and the way it is compiled, and the optimisation steps carried out to take maximum advantage of Bunya’s new hardware technologies.
“The internal network in Bunya is around four times faster than FlashLite or Tinaroo, and about double the speed of Wiener. You could transfer an entire 23 gigabyte Blu-Ray movie into a node on FlashLite or Tinaroo in around 3.28 seconds; Bunya can achieve the same transfer in 0.92 seconds,” Mr Carroll said.
“A practical example of the difference that internal high-speed interconnects can make to research might be for a machine learning researcher working on a data set of around 280 GB. To situate that in memory on FlashLite might have taken 40 seconds just to get the data into the node before training on the model could start. On Bunya, that transfer time could be reduced to around 11 seconds.”
The name ‘Bunya’ was chosen partly because of the native tree’s structural similarities to a supercomputer.
“The fruit of the Bunya is a tightly clustered super structure of individual segments making up a cohesive whole, just as supercomputers have interconnecting nodes to form a powerful whole,” Mr Carroll said.
Bunya will be available for UQ researchers and some QCIF member researchers.
Bunya is optimised for local users and campus infrastructure. It will start out as a traditional central processing unit (CPU)-based supercomputer with a number of high-memory capacity nodes for special uses.
Subsequent upgrades of Bunya will likely add different node types, such as high-performance accelerators, including GPUs.
The HPC will use novel methods of software deployment and management to allow for flexibility and ease of operation for researchers.
RCC’s Metropolitan Data Caching Infrastructure (MeDiCI) data fabric will allow research data collections to be accessed transparently via all UQ HPCs, including Bunya.
Bunya is currently being installed in the Polaris Data Centre in Springfield, Queensland, where Tinaroo and FlashLite are housed.
Media enquiries: Shannon Lindsay, Senior Communications Officer, UQ Research Computing Centre, s.lindsay@uq.edu.au.
