Aug. 3, 2020 — In parallel with the ongoing deployment of the JUWELS Booster, JSC is actively preparing its next major system upgrade. In November 2020, the JURECA Cluster will have reached five years of successful operation and will require renewal. For this purpose, the Cluster module of the modular supercomputer JURECA, consisting of the Cluster module as well as the Booster module installed in 2017, will be replaced by the new JURECA “data-centric” (DC) module.
The JURECA DC module will be supplied by Atos based on the Sequana XH2000 technology that is also used in the JUWELS Booster. The module will consist of a total of 768 compute nodes, all equipped with two 64-core AMD EPYC Rome CPUs and at least 512 GB of main memory. 96 of the nodes are equipped with 1 TB main memory. A quarter of the compute partition, 192 nodes in total, are equipped with four NVIDIA A100 GPUs. The system uses a HDR100/HDR InfiniBand interconnect with a scalable DragonFly+ topology. Compared to the predecessor, all major system performance indicators are increased. The core count of the system is increased by a factor of 4, the total memory capacity by 1.6 and the peak performance by approximately 7.8; the latter in particular due to the larger number of GPU nodes.
Like its predecessor, JURECA DC will be closely integrated with the JURECA Booster, enabling the concurrent use of all resources within workflows as well as the same parallel MPI application.
While JURECA DC will provide significant speedups for current JURECA workloads, several innovative features enable it to more efficiently execute data-intensive workloads. In particular, the JURECA DC module will include a novel system-integrated NVM in addition to high-bandwidth access to the JUST storage tiers.
The JURECA DC system will be deployed in two phases. The first phase will be available from the beginning of December and is planned to provide 384 CPU and 96 GPU compute nodes. The second phase will be operational in the first quarter of 2021.
The JURECA DC module was procured as part of the PPI4HPC project co-funded by the European Commission through a joint procurement process together with three European partner centres: BSC in Spain, GENCI/CEA in France, and CINECA in Italy. All four participating sites procured innovative HPC solutions for local and European users.
Computing time on the JURECA DC module will be made available to researchers of Forschungszentrum Jülich and RWTH Aachen University. Furthermore, a share of the compute time will be available for European scientists.
About Jülich Supercomputing Centre
The Jülich Supercomputing Centre at Forschungszentrum Jülich has been operating the first German supercomputing centre since 1987, and with the Jülich Institute for Advanced Simulation it is continuing the long tradition of scientific computing at Jülich. Computing time at the highest performance level is made available to researchers in Germany and Europe by means of an independent peer-review process. At the time being, JSC operates one of the most powerful supercomputers in Europe, JUWELS.
Source: Jülich Supercomputing Centre