Day of the Trifid: VPAC fires up new HPC cluster
$AU1.2m buy adds 45+ TFLOPS
The Victorian Partnership for Advanced Computing (VPAC), a consortium of Australian universities, has flipped the switch on 45.9 Teraflops of a new $1.22 million HP-based cluster to cope with rising workloads from partners La Trobe University and RMIT, and its other customers.
The Reg understands the 180-node, 2,880 core machine at VPAC is based on HP SL230 modules with DL380 standalone units as the login and management nodes.
According to Dr Mark Kosten, La Trobe University’s director of e-Research, workloads have been rising at all of Victoria’s supercomputing facilities. La Trobe’s cluster had hit 80 per cent usage, and both Monash’s MASSIVE and the University of Melbourne-based VLSCI are both showing similar growth.
“About two years ago, the director of eResearch at RMIT and I discussed joining forces on purchasing a new high-performance cluster, to gain performance not otherwise available to each of us alone,” Kosten told The Register in an e-mail.
Four nodes in the CentOS-based system are currently populated with NVIDIA Kepler GPUs, he said, with all nodes able to use GPUs (providing a straightforward upgrade path). Interconnects are provided by Mellanox Infiniband FDR switches over 10 Gbps Ehernet. The system includes 168 64 GB memory nodes and six 256 GB memory nodes.
Kosten also noted that the link between VPAC and La Trobe university, currently running at 1 Gbps, is due to be upgraded to 10 Gbps this year. ®
Sponsored: Network DDoS protection