Skip to content

Cluster description

Hardware

The Myria HPC cluster is an ATOS BULL solution with a power of 419 TFlops Xeon, 327 TFlops GPU and 27 TFlops Xeon Phi KNL:

  • 11304 CPU cores (10664 Xeon, 640 Xeon Phi KNL)
  • 366 Broadwell dual-socket compute nodes (28 cores @ 2,4 GHz, 128 GB of DDR4 RAM at 2400 MHz), including:
    • 12 nodes with 20 TB of internal disks each
    • 1 node with 20TB of internal disks and 1TB of RAM
    • 12 nodes each with 2 Kepler K80 GPU cards (4 GPUs per node, 12GB of on-board memory per unit)
    • 8 nodes each with 2 Pascal P100 GPU cards (2 GPUs per node, 12GB of on-board memory per unit)
    • 1 node with 3 Pascal P100 GPU cards (3 GPU processing units, 12 GB on-board memory per unit)
  • 5 SkyLake dual-socket computing nodes (16 cores @ 2,1 GHz, 187 GB of DDR4 RAM at 2666 MHz) each with 4 Volta V100-SXM2 GPUs (4 GPU processing units per node interconnected by NVLink2, 32 GB of onboard memory per unit)
  • 10 compute nodes each with a Xeon Phi KNL 7210 (64 cores @ 1,3 GHz and 96 GB of DDR4 RAM @ 2133 MHz by node, 16 GB of fast MCDRAM embedded in the KNL processor)
  • 1 SMP node with 256 Haswell cores at 2.2 GHz and 4 TB of DDR4 RAM at 2133 MHz
  • 2 visualization nodes (Broadwell dual-socket) each with a K80 GPU and 256GB of 2400MHz DDR4 RAM
  • 5 Broadwell dual-socket nodes, each with 256 GB of DDR4 RAM at 2400 MHz
  • 53 TB of DDR4 memory
  • Intel Omni-Path low-latency, high-speed (100 Gbps) interconnect network
  • 2,5 PB of shared disk space (/home, /dlocal et /save)
  • 4 x 10 Gbit/s and 1 x 40 Gbit/s connection on the SYVIK network

Software environment

  • 64-bit Linux environment (CentOS 7.6)
  • Submitting jobs with Slurm
  • IBM Spectrum Scale parallel file system (GPFS)
  • Development environment:
    • Intel 2017 and 2019 Compilers, Gnu 4.8.5 Fortran, C, C++ (OpenMP support)
    • Intel MPI 2017 and 2019 library (MPI-3 support)
    • Intel MKL math libraries (including BLAS, LAPACK and ScaLAPACK), FFTW 3.3.5
    • CUDA C/C++ API 8.0, 9.0, 9.1, 10.0, 10.1, 10.2 and 11.1, OpenACC 2.6 (Pgi 20.7), CUBLAS and CUFFT libraries for GPU acceleration
    • Makefiles example files: /soft/makefiles

Last update: November 25, 2022 14:05:21