Configuration

Hardware Configuration of the JURECA DC Module (Phase 2: as of May 2021)

  • 480 standard compute nodes
    • 2× AMD EPYC 7742, 2× 64 cores, 2.25 GHz

    • 512 (16× 32) GB DDR4, 3200 MHz

    • InfiniBand HDR100 (NVIDIA Mellanox Connect-X6)

    • diskless

  • 96 large-memory compute nodes
    • 2× AMD EPYC 7742, 2× 64 cores, 2.25 GHz

    • 1024 (16× 64) GB DDR4, 3200 MHz

    • InfiniBand HDR100 (NVIDIA Mellanox Connect-X6)

    • diskless

  • 192 accelerated compute nodes
    • 2× AMD EPYC 7742, 2× 64 cores, 2.25 GHz

    • 512 (16× 32) GB DDR4, 3200 MHz

    • 4× NVIDIA A100 GPU, 4× 40 GB HBM2e

    • 2× InfiniBand HDR (NVIDIA Mellanox Connect-X6)

    • diskless

  • 12 login nodes
    • 2× AMD EPYC 7742, 2× 64 cores, 2.25 GHz

    • 1024 (16× 64) GB DDR4, 3200 MHz

    • 2× NVIDIA Quadro RTX8000

    • InfiniBand HDR100 (NVIDIA Mellanox Connect-X6)

    • 100 Gigabit Ethernet external connection

  • 3.54 (CPU) + 14.98 (GPU) Petaflops per second peak performance

  • 98,304 CPU cores, 768 GPUs

  • Mellanox InfiniBand HDR (HDR100/HDR) DragonFly+ network
    • Ca. 15 Tb/s connection to Booster via gateway nodes

  • 350 GB/s network connection to JUST for storage access

Software Overview (as of Dec. 2018)

  • Rocky Linux 8 distribution (as of Dec. 2021)

  • Parastation Modulo

  • Slurm batch system with ParaStation resource management

  • DC module software stack
    • Intel Professional Fortran, C/C++ Compiler
      • Support for OpenMP programming model for intra-node parallelization

    • ParaTec ParaStation MPI (Message Passing Interface) implementation

    • Intel MPI

    • Open MPI

  • Intel Professional Fortran, C/C++ Compiler
    • Support for OpenMP programming model for intra-node parallelization

  • IBM Spectrum Scale (GPFS) parallel file system

Old Configuration Information

Hardware Configuration of the JURECA DC Module (Phase 1: 2020 - Apr. 2021)

  • 288 standard compute nodes
    • 2× AMD EPYC 7742, 2× 64 cores, 2.25 GHz

    • 512 (16× 32) GB DDR4, 3200 MHz

    • InfiniBand HDR100 (Connect-X6)

    • diskless

  • 96 large-memory compute nodes
    • 2× AMD EPYC 7742, 2× 64 cores, 2.25 GHz

    • 1024 (16× 64) GB DDR4, 3200 MHz

    • InfiniBand HDR100 (Connect-X6)

    • diskless

  • 48 accelerated compute nodes
    • 2× AMD EPYC 7742, 2× 64 cores, 2.25 GHz

    • 512 (16× 32) GB DDR4, 3200 MHz

    • 4× NVIDIA A100 GPU, 4× 40 GB HBM2e

    • 2× InfiniBand HDR (Connect-X6)

    • diskless

  • 12 login nodes
    • 2× AMD EPYC 7742, 2× 64 cores, 2.25 GHz

    • 1024 (16× 64) GB DDR4, 3200 MHz

    • 2× NVIDIA Quadro RTX8000

    • InfiniBand HDR100 (Connect-X6)

    • 100 Gigabit Ethernet external connection

  • 55,296 CPU cores, 192 GPUs

  • Mellanox InfiniBand HDR (HDR100/HDR) DragonFly+ network
    • Ca. 5 Tb/s connection to Booster via gateway nodes

  • 350 GB/s network connection to JUST for storage access

Hardware Configuration of the Cluster Module (2015 - 2020)

  • 1872 compute nodes
    • Two Intel Xeon E5-2680 v3 Haswell CPUs per node
      • 2× 12 cores, 2.5 GHz

      • Intel Hyperthreading Technology (Simultaneous Multithreading)

      • AVX 2.0 ISA extension

    • 75 compute nodes equipped with two NVIDIA K80 GPUs (four visible devices per node)
      • 2× 4992 CUDA cores

      • 2x 24 GiB GDDR5 memory

    • DDR4 memory technology (2133 MHz)
      • 1605 compute nodes with 128 GiB memory

      • 128 compute nodes with 256 GiB memory

      • 64 compute nodes with 512 GiB memory

  • 12 visualization nodes
    • Two Intel Xeon E5-2680 v3 Haswell CPUs per node

    • Two NVIDIA K40 GPUs per node
      • 2 x 12 GiB GDDR5 memory

    • 10 nodes with 512 GiB memory

    • 2 nodes with 1024 GiB memory

  • Login nodes with 256 GiB memory per node

  • 45,216 CPU cores

  • 1.8 (CPU) + 0.44 (GPU) Petaflop per second peak performance

  • Based on the T-Platforms V-class server architecture

  • Mellanox EDR InfiniBand high-speed network with non-blocking fat tree topology

  • 100 GiB per second storage connection to JUST for storage access

Hardware Configuration of the JURECA Booster Module (2017 - 09/2022)

  • 1640 compute nodes
    • 1× Intel Xeon Phi 7250-F Knights Landing, 68 cores, 1.4 GHz

    • Intel Hyperthreading Technology (Simultaneous Multithreading)

    • AVX-512 ISA extension

    • 96 GiB memory plus 16 GiB MCDRAM high-bandwidth memory

  • Shared login infrastructure with the JURECA-DC module

  • 111,520 CPU cores

  • 5 Petaflop per second peak performance

  • Intel Omni-Path Architecture high-speed network with non-blocking fat tree topology

  • 100+ GiB per second storage connection to JUST for storage access

Software Overview (until Dec. 2018)

  • CentOS 7 Linux distribution

  • Parastation Cluster Management

  • Slurm batch system with Parastation resource management

  • Intel Professional Fortran, C/C++ Compiler
    • Support for OpenMP programming model for intra-node parallelization

  • Intel Math Kernel Library

  • ParTec MPI (Message Passing Interface) Implementation

  • Intel MPI (Message Passing Interface) Implementation

  • IBM Spectrum Scale (GPFS) parallel filesystem

Booster software stack (Dec. 2018 - Sep. 2022)

  • Intel Professional Fortran, C/C++ Compiler

  • Intel Math Kernel Library

  • ParTec Parastation MPI

  • Intel MPI