The Australian government has committed $70 million in funding for a next-generation number cruncher at the Pawsey Supercomputing Centre, located in Perth, Western Australia. The investment will enable the tier-1 center to replace its flagship supercomputer, Magnus, as well as the real-time supercomputer, Galaxy — both of which are nearing end-of-life.
Magnus is a 1.1-petaflops (Linpack) Cray XC40 machine that is one of the most advanced supercomputers in the southern hemisphere, facilitating research in medical science, engineering, geoscience, marine science, chemistry, food, agriculture and more. Companion system Galaxy supports the operational requirements of the Australian Square Kilometre Array (SKA) pathfinder telescopes.
Last December, the Australian government announced an equal funding award ($70 million) to replace Australia’s highest performance supercomputer, Raijin, at the National Computational Infrastructure National Facility (NCI-NF) in Canberra (the country’s other tier-1 center). A combined Fujitsu-Lenovo system powered by Xeon Broadwell CPUs, Nvidia P100 GPUs and Xeon Phi KNL processors, Raijin delivers 1.68 petaflops Linpack (3.8-petaflops peak), supporting researchers in 35 universities, five national science agencies, three medical research institutes, and industry. Since its 2013 debut, Raijin has kept its post as Australia’s fastest supercomputer thanks to upgrades in 2016 and 2017 (the latter for four IBM Power8 nodes); it is currently ranked 76th on the Top500 listing of fastest computers.
Mr. John Langoulant, chairman of the Pawsey Supercomputing Centre Board, commented that this new award together with last year’s investment into the National Computational Infrastructure “will strengthen Australia’s position in the global research environment and enable Australia to stay globally competitive.”
With several of Australia’s large high-performance computers nearing the end of their service, the funding commitments are essential to ensuring Australia meets its research goals.
On the occasion of the NCI funding announcement last December, Australia’s Chief Scientist Dr. Alan Finkel upheld high-performance computing as a national priority, referencing Australia’s 2016 National Research Infrastructure Roadmap.
“An immediate priority is the need to refresh Australia’s national HPC,” the government report stated. “This should be coupled with a review of existing governance arrangements to maximise the strategic position and accessibility of national HPC.”
The procurement process at Pawsey will proceed immediately with the intention to launch a new system in 2019. NCI’s tender process is already underway and it is also expecting to commission a new system in 2019.
Top Australian science agency CSIRO, a partner of both tier-1 HPC facilities in Australia, spent $4 million last year on a new Dell EMC supercomputer, dubbed “Bracewell” in honor of Australian astronomer and engineer Ronald N. Bracewell. With 456 Nvidia P100 GPUs providing nearly 2.4 peak petaflops of performance, the system doubles the computational power available to CSIRO researchers and brings deep learning capability to the agency. It is a key resource for the CData61 Computer Vision research group.
Installed at Canberra Data Centres last June, Bracewell boasts 114 PowerEdge C4130 servers with dual 14-core Intel Xeon processors and four NVLink’d Nvidia P100 GPUs, tied together with EDR InfiniBand interconnect. Total system memory: 29 TB. Based on raw power, the machine could expect to place in the middle of the Top500 pack, however despite a summer 2017 deployment, it did not appear on the November list.