Upcoming HPC Systems in 2020

In order to provide a world-class infrastructure to its users and drive the development of high-performance computing and storage technologies for scientific cases forward, the supercomputing facility at JSC is undergoing continuous rework and upgrades. The year 2020 will be a particularly busy year.

In spring 2020, a new storage system within the Jülich Storage cluster (JUST) will be made available. The new high-performance tier leverages non-volatile memory (flash) technology to provide very high bandwidth and accelerate a range of I/O-intense simulation and machine learning workloads with the “Infinite Memory Engine” technology by DDN. The system integrator is HPE. The storage system will provide separate storage pools to JUWELS, the JURECA Cluster, and JUSUF (see below) with an accumulated bandwidth of more than 2 TB/s. In a joint co-development effort, DDN and JSC are working on enabling coherent access between the storage slices attached to the three computing systems to enable workflows across the entire modular supercomputing facility at JSC. This capability is expected to be available in 2021.

Also in early 2020, the new JUSUF system, short for Jülich Support for Fenix, will go into production. JUSUF, which is delivered to JSC by Atos, combines an HPC cluster and a cloud platform in a single system with homogeneous hardware such that resources can be flexibly shifted between the partitions. The JUSUF compute nodes are equipped with two AMD EPYC Rome CPUs, each with 64 cores operating at 2.2 GHz. One third of the compute nodes are furthermore equipped with one NVIDIA V100 GPU. The JUSUF cluster partition will provide HPC resources for interactive workloads and batch jobs. The cloud partition will enable co-location of (web) services with these resources to enable new workflows and support community platforms.

The JUST high-performance storage and JUSUF are co-funded by the European ICEI project, which is establishing the new federated e-infrastructure “Fenix”. A share of the resources will be made available to the Human Brain Project community and to European researchers via PRACE. More information about resource access to these infrastructure components will be published soon.

Later in 2020, JSC will install the new Booster module for the Tier-0/1 system JUWELS. The module will boost the peak performance of JUWELS to more than 70 petaflops. The JUWELS Booster compute nodes are equipped with multiple next-generation NVIDIA GPUs and Mellanox 200 Gb/s HDR InfiniBand adapters. The JUWELS Cluster and Booster are integrated in the same high-speed interconnect fabric enabling applications to efficiently leverage the modular architecture of JUWELS.

Finally, in late autumn, the JURECA Cluster module will have reached five years of successful operation and is scheduled for decommissioning. The procurement of the successor system will be finalized by February 2020. Due to infrastructure constraints, parallel operation of the successor and the JURECA Cluster will not be possible. Therefore, JSC intends to perform installation in phases to minimize service interruption. More details about the time schedule and the architecture of the successor system will be published in due time. The system is co-financed by the European Commission in the context of the PPI4HPC project. A share of its resources will be available to European researchers.

JSC is committed to minimizing and, where possible, eliminating service disruptions due to these deployment activities. However, in some cases additional maintenance will be required. Our thanks go to our users for their patience and continued support.

Contact: Dr. Dorian Krause, d.krause@fz-juelich.de

from JSC News No. 269, 23 January 2020

Last Modified: 11.03.2022