CSIRO seeks $14.5 million Bracewell supercomputer replacement

CSIRO has launched a $14.5 million tender for a supercomputer to replace its five-year-old Bracewell machine.

As well as traditional HPC workloads, the Advanced Scientific Accelerator Cluster (ASAC) will have to be able to handle machine learning and artificial intelligence workloads.

CSIRO gives astronomy, manufacturing, climate sciences, health and biosecurity as examples of the research workloads the ASAC will need to run.

It will be installed in the CSIRO’s floor space at Canberra Data Centres.

The tender covers servers with a high-performance interconnect, an out-of-band management network for all system nodes, software including a “comprehensive application development environment”, along with cluster management software and monitoring tools.

Software-defined networking support is required to “segregate/optimise specific workflows”, the tender said.

While not stipulating the scale of the GPU accelerator compute nodes beyond 80GB memory per node, the tender states there will be three interactive nodes and one data mover node, each of which needs 2TB of memory; while x86_64 architecture is specified for data visualisation nodes.

Due by September 2023, the system will have to fit in the 14 racks available to CSIRO in CDC.

Its predecessor Bracewell went live in July 2017. It’s “a homogeneous cluster consisting of 114 Dell PowerEdge C4130 servers with dual sockets, quad P100 GPUs, 256GB of memory and 960GB local disk with EDR interconnect”, the tender states.

The $4 million machine was named after Australian astronomer and engineer Ronald N Bracewell, who helped to advance medical imaging technology.