News Articles

HiPerGator Software Environment Update

📅 May 17, 2024

What is happening?

All HiPerGator systems are being upgraded from Red Hat Enterprise Linux version 8.8 to version 8.9, including all security patches, bug fixes, and software enhancements. User-visible third party software used on HiPerGator systems is being updated to the latest generally available versions from upstream sources (e.g., NVIDIA, OpenMPI, and PMIx communities, etc.) Corresponding new modules will be made available. HiPerGator systems are also being updated to the latest generally available versions of System firmware from our hardware vendors.

When is it happening?

These updates will be performed in a rolling fashion across all resources beginning on 13 May 2024 . They are expected to complete across all resources before 21 Jun 2024 . Any schedule adjustments (if needed) will be communicated via the HPC-Announce mailing list, UFIT Research Computing announcements Slack channel, and corresponding news articles posted on the rc.ufl.edu website.

What do you need to know?

The system updates are expected to be largely invisible to workflows executed on HiPerGator. Highlights of the update are as follows:

  • CUDA Toolkit: version 12.2 Update 2 is to be updated to 12.4 Update 1 on HPG-AI and general GPU partitions. Numerous bug fixes and performance improvements.
  • PMIx: Version 3.2.5 is being deployed and is a security update. The existing PMIx 3.2.4, which openmpi/4.1.5 modules are linked against, will remain available on HiPerGator until further notice. This version is now deprecated and will ultimately be removed. Future communications will detail the removal of version 3.2.4.
  • Unified Communication X (UCX) software from the OpenUCX collaboration is being updated on all systems. This is primarily of interest to users who use MPI and the versions used are platform-dependent:
    • HiPerGator2: version 1.13.1 will be updated to 1.14.1.
    • All other resources, including HiPerGator3, HiPerGator-AI, and most hosts in the GPU and hwGUI partions: version 1.15.0 will be updated to 1.16.0.

The following new modules have been made available to take advantage of the updates:

  • cuda/12.4.1
  • openmpi/4.1.6 - this is a bugfix release for OpenMPI, but is linked against the updated PMIx noted above.
    • Users who build their own applications using the openmpi/4.1.5 are strongly encouraged to recompile and relink their executables using the new openmpi/4.1.6 module immediately

Individual login hosts will be rebooted as part of the update process. This will occur in a rolling fashion during the evenings of May 20, 2024, May 21, 2024, and May 22, 2024. The load-balanced HiPerGator login services themselves will remain available at all times, and will not be affected by the individual host reboots.

Problems or questions?

Please open a support request at support.rc.ufl.edu for any concerns, problems, or questions regarding the above.