NCI Doubles Raijin Supercomputer Throughput with IBM POWER8

Raijin Supercomputer

Today NCI in Australia announced that it has adopted IBM’s Power8 architecture as part of Raijin, the system  fastest supercomputer in the Southern Hemisphere. The hybrid x86/Power8 system will offer local researchers the opportunity to explore the intersection of AI and HPC.

IBM’s Power Systems scale scientific simulation and modeling to new heights in NCI’s Raijin supercomputer,” said Dave Turek, Vice President of Exascale Systems at IBM. “The extraordinary bandwidth in Power Systems provides a significant performance advantage, and we look forward to scientists exploiting those capabilities now and into the future.”

NCI staff have been working closely with Australia’s scientific community to qualify a range of memory-intensive applications for use under IBM’s architecture. With further input from NCI’s optimization team, these applications have produced impressive performance benchmarks using Power Systems.

For example, NCI is the first institution to port Q-Chem, a quantum chemistry package, over to Power Systems. Initial benchmarks of this optimization have outpaced the same application running under Broadwell x86 architecture, opening new and immediate possibilities for computational chemists.

Additionally, NCI’s optimization of NAMD molecular dynamics code on Power Systems saw performance gains when compared to existing Broadwell x86 architecture. Initial benchmark results of MILC in Power Systems also saw modest to significant improvements over existing x86 systems.

Based on the benchmarks from the research testbed, it is anticipated that a wide range of scientific disciplines could eventually benefit from optimized performance under POWER architecture, including (but not limited to) the fields of physics, biology and chemistry.

This new software exists alongside over 3,000 other applications that have already been optimized by NCI’s HPC optimization team for the existing x86 and GPU systems that power Raijin. NCI will continue a process of Power Systems qualification and optimization for other scientific applications to meet researcher demand.

NCI is the first institution to merge both POWER and x86 architectures into the same scheduling system, offering unparalleled diversity, accessibility and freedom of choice for researchers accessing HPC through NCI. The IBM Power Systems are accessible through the existing PBSPro scheduler, and have been integrated into NCI’s LDAP and accounting systems.

Furthermore, the Power Systems will benefit from Mellanox EDR and FDR interconnect at speeds of up to 100Gbps, granting access to over 40PB of fast Lustre storage that is already available across other Raijin platforms.

NCI’s commitment to investing in the latest computing resources continues with the integration of IBM’s Power Systems into the Raijin cluster. This integration builds on NCI’s consolidated approach to advanced computing, combining high-performance CPU and GPU clusters with the fastest filesystems in the southern hemisphere – a heterogenous environment that will best serve the research community’s needs.

Sign up for our insideHPC Newsletter