Search

link to homepage

Institute for Advanced Simulation (IAS)

Navigation and service


IBM iDataPlex Cluster JUDGE

JUDGE_multi_image

Compute Nodes

  • 206 Compute nodes IBM System x iDataPlex dx360 M3

    • Compute node: 2 Intel Xeon X5650(Westmere) 6-core processor 2.66 GHz
    • Main memory: 96 GB
    • Network: IB QDR HBA
    • GPU: 2 NVIDIA Tesla M2050 (Fermi) 1.15 GHz (448 cores), 3 GB memory (on 54 Nodes)
    • GPU: 2 NVIDIA Tesla M2070 (Fermi) 1.15 GHz (448 cores), 6 GB memory (on 152 Nodes)

Complete System

  • 2472 cores total
  • 412 graphic processors
  • 19.8 TB main memory
  • 239 Teraflops peak performance

Infrastructure Nodes

  • GPFS gateway nodes

    • 2 IBM System x iDataPlex dx360 M3 ( 8 cores, 12 GB memory)
  • Login node and management nodes

    • 4 IBM System x iDataPlex dx360 M3 (12 cores, 96 GB memory)
    • Internet address: judge.fz-juelich.de
    • Internet address: milkyway.fz-juelich.de

Filesystem

  • GPFS file systems from JUST

Software

  • Operating system: SUSE SLES 11
  • Cluster management: ParaStation

    • GridMonitor
  • Operating mode: Torque/Maui Batch and Resource Management System
  • Compiler: Intel Professional Fortran, C/C++
  • Intel Cluster Tools

    • Intel Math Kernel Library
  • ParTec MPI Message Passing Interface
  • OpenMP Intra-Node Programming Model

In February 2011 production of the iDataplex Cluster JUDGE (JUelich Dedicated Gpu Environment) started with 54 nodes. It is available for IAS users and JSC projects.

In November 2011 the system was enlarged to 122 nodes. The new nodes are dedicated to JuBrain, a common platfrom for simulations on neuroscience of INM and JSC.

In Spring 2012 the new Milkyway partion was added to judge. The new partition was designed and purchased in a common effort involving JSC, the university on Heidelberg within the DFG Sonderforschungsbereich 881 "The Milky Way System" and the Ludwig-Maximimillan-Universität München within the DFG Schwerpunktprogramm 1553 "Physics of the Interstellar Medium". The "Milky Way" part consists of 84 compute nodes with own login an d GPFS gateway nodes. Large jobs can run on all 206 compute nodes of JUDGE as both parts are coupled via Infiniband connections. There is a special queue schema for users of the "Milky Way" part.

Further information about allocation of computing time and how to get an account can be found at


Servicemeu

Homepage