Extension of the Jean Zay supercomputer to 28 Pflop/s

The company GENCI (Grand Equipement National De Calcul Intensif), in coordination with the CNRS and with the support of the IDRIS teams, has launched the extension procedure of the Jean Zay supercomputer with Hewlett Packard Enterprise (HPE). The decision was made to extend the peak performance of Jean Zay to 28 Pflop/s, thereby making it one of the most powerful accelerated hybrid supercomputers in Europe.

This extension will be in HPE SGI 8600 technology similar to that of the initial phase, providing 351 supplementary accelerated nodes consisting of:

  • 2 Intel Cascade Lake 6248 processors (20 cores at 2.5 GHz), or 40 cores per node.
  • 192 GB of memory per node.
  • 4 GPU Nvidia Tesla V100 SXM2 16 GB (vs. 32 GB for the initial phase) per node.
  • The extension of the Intel Omni-Path fabric of the supercomputer.
  • The extension of the SSD parallel storage device (DDN GS18K) which will increase to 2.2 PB.

The extension will increment the power by 12 Pflop/s (1404 supplementary GPUs) giving a final configuration of 28 Pflop/s and a total of 2696 V100 GPUs.

The technical description of the Jean Zay configuration with the extension can be found on the IDRIS Web site on the following page:

http://www.idris.fr/eng/jean-zay/cpu/jean-zay-cpu-hw-eng.html

The extension will be installed this summer and will be open to all users for the A9 call of October 2020.