Dear HPC Users,

Good news: acceptance tests of the new "LEO5" compute cluster are in their final stage. We expect to start "friendly user" test operation on February 01, 2023.

Before going into details, we belatedly wish to thank all participants of our users' survey at the beginning of this project. Your input was of exceptional quality and has had substantial influence on our specifications for the LEO5 tender. In particular, we believe that most of your resource requirements (e.g. memory, storage, GPUs) will be met by LEO5. However, shared memory needs beyond 2 TB per host had to be kept outside the scope of this project.

The total performance of LEO5 sets a new standard in the LEO lineup: the nominal floating point CPU performance is estimated to approx. 200 TFlop/s (HPL). The total GPU performance is nominally 600 TFlop/s (Single), 220 TFlop/s (Double Precision).

For details, see

LEO5 key features
Hardware
  • 62 compute nodes with two 32-core Intel Xeon 8358 sockets each (i.e. 64 cores/node);
    • 57 nodes with 256 GB memory each, 5 with 2 TB;
    • 21 nodes with two Nvidia A30 GPUs, 2 with two A40 GPUs.
  • The login node has 2 TB  memory and 2 A30 GPUs.
  • Nodes are homogeneously networked with Infiniband HDR100 (no islands),
  • Currently projected net storage capacity is 2.1 PB (120 TB SSDs)
Software
  • Rocky Linux 8.6 (RedHat Enterprise Linux clone)
  • GPFS high performance parallel file system
  • Load manager: Slurm (replaces obsolete SGE)
  • Software deployment systems: Spack and (for Python and R) Conda
Access
All existing LEO accounts will be activated for LEO5 on February 01, 2023.
Login via
  ssh cXXXyyyy@leo5.uibk.ac.at

"Friendly Users Test Operation" means in particular:
  • Documentation is in progress,
  • System operation may be interrupted at any time, possibly without prior notice,
  • Software availability may be limited or subject to changes,
  • Your friendly feedback helps us discover errors and improve the configuration.
What happens to the existing clusters
  • LEO3e and LEO4 operation will continue. Central components are covered by maintenance contracts until March 2024 (LEO3e) and November 2024 (LEO4).
  • LEO3 has been dismantled. Part of its hardware will replace the LCC2 teaching cluster (LCC3).
  • End of Life of MACH2 is pending - last day of access is scheduled for March 31, 2023. Active users will be informed in detail via email shortly.
Regular operation of LEO5 is planned to start in May 2023.

We are looking forward to another chapter of successful service and cooperation with our HPC community. Please to not hesitate to contact us if you need advice or support using our systems.

Sincerely, your HPC Team.

Nach oben scrollen