Cardiff University expands HPC cluster with Lenovo

Cardiff University has deployed 90 Lenovo ThinkSystem servers, with Logicalis technology, to support high-performance computing (HPC) research. The university’s Advanced Research Computing @Cardiff (ARCCA) service worked with Logicalis and the Lenovo ThinkSystem platform to boost the capabilities of its Hawk HPC cluster by up to 100%. The HPC will be used to support research into the origins of the universe and star formation.

Lenovo has provided 75 Lenovo ThinkSystem SR645 servers ahead of the fourth two-year gravitational wave detection run of the Laser Interferometer Gravitational-Wave Observatory (LIGO) Scientific Collaboration. The servers offer a minimum performance boost of 46% compared with a traditional HPC environment using identical hardware. According to Lenovo, for some applications, the new servers double the compute performance available to researchers. This uplift in performance enables researchers to process detection events and share data faster.

The hardware comprises Lenovo ThinkSystem SR645 servers powered by AMD Epyc processors, which connect to the Hawk cluster using Nvidia ConnectX-6 connectivity via Ubiquity, an open source cluster management tool provided by Logicalis.

Stephen Fairhurst, head of the gravity exploration institute at Cardiff University, said: “We are getting even greater per-core performance than we expected. Crucially, the new cluster was up and running in time for the latest LIGO observing run. We can now process detection events faster and can share the data with astronomers sooner, helping them locate sources of the signals more easily.”

A second research group at Cardiff University studying star formation, led by Royal Society University research fellow Ana Duarte Cabral, will use 15 Lenovo ThinkSystem SR630 servers with two Lenovo ThinkSystem SR650 servers for storage. Funded by the Royal Society, the group’s research focuses on the creation of stars in spiral galaxies like our own Milky Way, requiring computationally demanding simulations of full galaxies.

“We are getting even greater per-core performance than we expected. We can now process detection events faster and can share the data with astronomers sooner, helping them locate sources of the signals more easily”