Compute nodes phase 1

Hardware of Phase I (operating since 2013)

4 login nodes

  • 4x processors “Intel® Xeon® processor E5-4650”
    • Each processor consists of 8 cores
    • Hyperthreading is off
    • There is an AVX unit per core
  • Overall 32 cores per node
  • Processor clock is 2,7 GHz – with a turbo up to 3,3 GHz (if you use less cores)
  • node memory expansion: 128 GByte per node
  • network connection: 1x FDR-10 InfiniBand and 2x 10GBit-Ethernet
  • Name of the system: hla0001 … hla0004
  • Accessible from outside with: lcluster1 … lcluster4

MPI – Section (706 nodes)

  • 2x processors “Intel® Xeon® processor E5-2670”
    • Each processor consists of 8 cores
    • Hyperthreading is off
    • There is a AVX unit per core
  • Overall 16 cores per node
  • Processor clock is 2,6 GHz – with a turbo up to 3,3 GHz (if you use less cores)
  • Node memory expansion: 32 GByte per node, island-16 and -17 with 64 GByte per node
  • Network connection: 1x FDR-10 InfiniBand and 1x 1GBit-Ethernet
    • Within the island all nodes are connected with a blocking relation of 1:1
    • The cross-island blocking relation is approximately 1:8
  • Name of the system: hpa0001 … hpa0706

MEM – Section (4 nodes)

  • 8x processors “Intel® Xeon® processor E7-8837”
    • Each processor consists of 8 cores
    • Hyperthreading is deactivated
    • There is no AVX unit
  • Overall 64 cores per node
  • Processor clock is 2,66 GHz – with a turbo up to 2,8 GHz
  • Node memory expansion: 1024 GByte
  • Network connection: 2x FDR-10 InfiniBand and 1x 10GBit-Ethernet
  • Name of the system: hea0001 … hea0004

ACC – Section (70 nodes)

  • 2x processors “Intel® Xeon® processor E5-2670”
  • Processors, memory and network simiilar to MPI – Section
  • 2x accelerator cards per node:
    • NVD: 44x nodes, each with two “NVIDIA® Tesla™ K20X”
      • each approx. 1,31 TFlop/s computing power (DP-exactness peak – theoretical)
        • 2688 CUDA units
        • 6 GByte local memory
        • 250 GByte/s memory bandwith (peak – theoretical)
      • Name of the system: han0001 … han0044
    • PHI: 24x nodes each with two “Intel® Xeon Phi™ coprocessor 5110P” (deactivated since August 2018)
      • each approx. 1,0 TFlop/s computing power (DP-exactness peak – theoretical)
        • 60 cores (1,05 GHz)
        • 8 GByte local memory
        • 320 GByte/s memory bandwith (peak – theoretical)
      • Name of the system: hai0001 … hai0024
    • PHI: 2x nodes each with two “Intel® Xeon Phi™ coprocessor 7120P” (deactivated since August 2018)
      • each approx. 1,2 TFlop/s computing power (DP-exactness peak – theoretical)
        • 61 cores (1,24 GHz, turbo 1.33 GHz)
        • 16 GByte local memory
        • 352 GByte/s memory bandwith (peak – theoretical)
      • Name of the system:hai0025 … hai0026