JURECA

The modular supercomputer JURECA is operated by the Jülich Supercomputing Centre and consists of the JURECA Booster Module and the JURECA-DC Module.

 

The JURECA Booster Module

The JURECA Booster Module has a peak performance of 5 PFlop/s. It is available for all JARA researches.

Node characteristics

Available resources per call

  • One Intel Xeon Phi 7250-F Knights Landing CPU
  • 68 cores, 1.4 GHz
  • 96 GB per node  (~ 1.4 GB per core) +
    16 GB MCDRAM high-bandwidth memory
  • About 1640 nodes are available

180 million core-h*) or about 29000EFLOP*)

*) Contains about 100 million core-h (16000EFLOP) dedicated to applicants from FZJ

 

The JURECA-DC Module

The new JURECA-DC (data-centric) Module consists of compute nodes with GPUs (GPU nodes) and without GPUs (CPU nodes).  

Node type

Node characteristics

Available resources per call

CPU nodes
  • Two AMD EPYC Rome 7742 CPUs
  • 2 x 64 cores, 2.25 GHz per node
  • at least 512 GB main memory per node (~4 GB per core)

135 million core-h or about 17500 EFLOP

GPU nodes
  • Two AMD EPYC Rome 7742 CPUs
  • 2 x 64 cores, 2.25 GHz
  • 512 GB main memory per node (~4 GB per core)
  • Four NVIDIA A100 GPUs
  • 4 x 40 GB HBM2e memory
  • 192 nodes are available
45 million core-h or about 104500 EFLOP**)

**) Resources on GPU nodes are accounted in core-h of the host CPU, i.e. 128 core-h on JURECA-DC GPU nodes means using 1 node with four A100 GPUs for 1 hour.

IMPORTANT NOTE: Phase 2 of the JURECA‐DC (data centric) module is now in operation at JSC and available mainly for scientists from the FZJ campus. However, researchers of the RWTH Aachen university can also apply for resources on JURECA‐DC CPUs and/or GPUs within JARA, if they benefit from the Modular Supercomputing Architecture (MSA) of the system. Requirements for this are:

  • A need for at least two resources/modules is required, i.e., either JURECA‐DC CPUs together with JURECA‐DC GPUs or JURECA‐DC CPUs together with the JURECA Booster Module.
  • Either heterogeneous applications have to be used, where, for example, different parts of a simulation code use different physical models and run therefore on different resources/modules. Or heterogeneous work‐flows are deployed, for example, to model phenomena on different length scales or to perform chains of simulation and data analysis parts, where the different parts run on different resources/modules.
  • The modularity of applications and/or work‐flows must be clearly described in the application form.

 
JSC offers data projects in addition to computing time projects. For a fact sheet on data projects and the application form for data projects please see here. For further information on this please contact the user support (sc@fz-juelich.de) at JSC.