Genius hardware

Genius is one of KU Leuven/UHasselt’s Tier-2 clusters, which is in production since 2018. It has thin nodes, large memory nodes, as well as GPU nodes.

Note

Most of Genius will be decommissioned in early 2026.

Hardware details

• 144 thin nodes

  • 2 Xeon Gold 6240 CPUs@2.6 GHz (Cascadelake), 18 cores each
    (1 NUMA domain and 1 L3 cache per CPU)

  • 192 GiB RAM (memory bandwidth and latency measurements)

  • default memory per core is 5000 MiB

  • 200 GB SSD local disk

  • partition batch|batch_long, specific Slurm options apply

• 12 interactive nodes

  • 2 Xeon Gold 6140 CPUs@2.3 GHz (Skylake), 18 cores each
    (1 NUMA domain and 1 L3 cache per CPU)

  • 192 GiB RAM (memory bandwidth and latency measurements)

  • default memory per core is 2000 MiB

  • 200 GB SSD local disk

  • partition interactive, specific Slurm options apply

• 10 big memory nodes

  • 2 Xeon Gold 6140 CPUs@2.3 GHz (Skylake), 18 cores each
    (1 NUMA domain and 1 L3 cache per CPU)

  • 768 GiB RAM

  • default memory per core is 21000 MiB

  • 200 GB SSD local disk

  • partition bigmem|bigmem_long, specific Slurm options apply

• 22 GPGPU nodes, 96 GPU devices

  • 20 P100 nodes

    • 2 Xeon Gold 6140 CPUs@2.3 GHz (Skylake), 18 cores each
      (1 NUMA domain and 1 L3 cache per CPU)

    • 192 GiB RAM

    • default memory per core is 5000 MiB

    • 4 NVIDIA P100 SXM2@1.3 GHz, 16 GiB GDDR, connected with NVLink

    • 200 GB SSD local disk

    • partition gpu_p100|gpu_p100_long|gpu_p100_debug, specific Slurm options apply

  • 2 V100 nodes

    • 2 Xeon Gold 6240 CPUs@2.6 GHz (Cascadelake), 18 cores each
      (1 NUMA domain and 1 L3 cache per CPU)

    • 768 GiB RAM

    • default memory per core is 21000 MiB

    • 8 NVIDIA V100 SXM2@1.5 GHz, 32 GiB GDDR, connected with NVLink

    • 200 GB SSD local disk

    • partition gpu_v100|gpu_v100_long, specific Slurm options apply

The nodes are connected using an Infiniband EDR network (bandwidth 25 Gb/s), the islands are indicated on the diagram below.