Full Availability of JURECA-DC Module

The JURECA-DC module – DC for data-centric – is a supercomputer with innovative support for data-intensive applications. Its first phase with 384 CPU nodes, each with 128 AMD Rome EPYC cores per node, and 48 nodes additionally equipped with four NVIDIA A100 GPUs, entered production in December 2020 as the successor of the JURECA Cluster module.

On 1 May 2021, the second phase of the JURECA-DC module was brought online – right in time for the start of the new computing-time period. The system now consists of 480 CPU nodes of the aforementioned CPU type with 512 GB memory each, 96 CPU nodes with 1024 GB memory each, as well as 192 GPU nodes with 512 GB of memory and four A100 GPUs each. The compute nodes and login servers are interconnected by a Mellanox InfiniBand HDR network. User access is possible through 12 login servers.

The theoretical peak performance is now 18.52 petaflops, with 14.98 petaflops on the GPUs and 3.54 petaflops on the CPUs. In its final configuration, the JURECA-DC module has 8 times the performance of its predecessor, the JURECA Cluster module.

Within the JURECA-DC module, additional features are available, like a cluster-local network-accessible NVMe-based storage system, additional profiling capabilities, and an energy optimizer. Further information on JURECA-DC, its configuration, and the available tools can be found at https://fz-juelich.de/ias/jsc/jureca.

The JURECA-DC module was procured as part of the EU-funded PPI4HPC project, in which computing centres from four European countries have united to purchase new, innovative supercomputer systems through a joint procedure – for the first time at the European level. A share of its resources will therefore be available to European researchers. Predominantly, computing time on JURECA-DC will be granted to researchers at Helmholtz centres, and in particular to researchers at Forschungszentrum Jülich.

Contact: Philipp Thörnig, p.thoernig@fz-juelich.de

from JSC News No. 281, 16 June 2021