The National Computational Infrastructure (NCI) has picked the new kit that will help boost the capacity of its Raijin high performance compute (HPC) system by 40 percent.
Its new Lenovo NeXtScale cluster will be supplied by Xenon Systems and is expected to enter production during January next year, NCI said today.
The system contains 22,792 Intel Xeon Broadwell 2690v4 processors and 144 terabytes of memory, including 10 one-terabyte nodes.
The nodes will be connected using Mellanox EDR 100Gbit/s Infiniband interconnects, configured as a fat tree, and will be implemented as an extension of the organisation’s existing Raijin supercomputer.
The new nodes will be integrated with Raijin’s existing 7.6 petabytes of local Lustre scratch storage accessible at 150 gigabytes per second, with access to a further 25 petabytes of global Lustre storage.
The upgrade is in addition to the 14 nodes NCI added to Raijin at the end of May. The system currently boasts a total of 3609 compute nodes and 57, 864 compute cores, giving it a total 1.37 petaflops of total compute power.
The upgrade has been funded as part of a $7 million grant from the National Collaborative Research Infrastructure Strategy (NCRIS) Agility Fund, awarded in July, which was matched by an additional $7 million grant from partner research agencies.
Aside from the additional compute capacity, the grants will also pay for a 3 petabyte increase in the storage capacity with the replacement of NCI’s oldest Lustre storage system, expected to take place next year.