University of Hamburg - to research, to teach, to educate and form, to homepage
RRZ
Regional Computing Center
(RRZ)

Photo: Thomas Orgis

hardware

Compute nodes

All compute nodes are diskless. All CPUs, memory modules and GPUs are directly water cooled. The compute nodes can only be accessed via the batch system. Hummel-2 has:

  • 172 standard CPU nodes
    • 2 AMD EPYC 9654 CPUs (2 × 96 cores)
    • 768 GB RAM
    • partition: std
    • node names: n001..n172
  • 6 big CPU nodes
    • 2 AMD EPYC 9654 CPUs (2 × 96 cores)
    • 2304 GB RAM
    • partition: big
    • node names: b001..b006
  • 4 GPU nodes
    • 8 NVIDIA H100 80GB GPUs
    • 2 AMD EPYC 9334 CPUs (2 × 32 cores)
    • 1152 GB RAM
    • partition: gpu
    • node names: g001..g004

Front-end nodes

The front-end nodes are used for interactive work (i.e. preparing and managing batch jobs). On many clusters these kinds of nodes are called login nodes. However, on Hummel-2 the front-end nodes can only be access from login gateway nodes. There are:

  • 2 front-end nodes
    • 1 AMD EPYC 9354 CPU (32 cores)
    • 192 GB RAM
    • node names: front1, front2

Login gateway nodes

The login gateway nodes are the access points to Hummel-2 from the internet. There are:

  • 2 login gateway nodes
    • the hostname of node login1 is hummel3.rrz.uni-hamburg.de
    • the hostname of node login2 is hummel4.rrz.uni-hamburg.de
    • fingerprints of host keys are given on the Quick Start page

Disk systems

Note: There is no backup, except for the /home file system!
Please read also Working with data.

  • HDD (spinning) disk systems
    All file systems on HDD are based on the ZFS file system configured with dRAID.

    • Parallel file system /beegfs
      • BeeGFS file system
      • backup: no
      • capacity: 3.2 PB
    • NFS file systems
      • used by large data projects
      • backup: no
      • capacity: 4 × 540 TB

  • SSD/NVMe disk systems

    • Home file system /home

    • User-installed software /usw

    • SSD/NVMe disk pool

      • 33 × 15 TB SSDs that can be flexibly assigned to compute nodes for data-intensive computing tasks either via NFS or NVMe-oF
      • no RAID
      • backup: no

Interconnect