Is this page helpful?

Release notes for NVIDIA Base Command™ Manager (BCM) 10.31.0#

Released: May 7 2026

General#

New Features#

  • Added CUDA 13.1 packages

  • Updated Slurm 24.11 to 24.11.7

  • Updated Slurm 25.05 to 25.05.6

  • Updated Slurm 25.11 to 25.11.5

  • Updated munge to 0.5.18 (CVE-2026-25506)

  • Updated Ubuntu 24.04 base OS to 24.04.3

  • Updated cm-nvhpc to 26.1

  • Updated cm-openssl to 3.5.5

  • Updated cuda-driver to 580.126.20

  • Updated cuda13.0 to 13.0.2

  • Updated cuda13.1 to 13.1.1

  • Updated golang-go-latest to version 1.25.6

  • Updated lib-prometheus to 1.26.1

  • Updated cm-nvidia-container-toolkit to v1.18.2

  • Kubernetes deployments now use Calico helm charts (Tigera operator)

Fixed Issues#

  • An issue in the enroot prolog that did not update permissions for /var/lib/enroot

  • An issue where workload manager epilog and prolog scripts were not running in the correct order

  • An issue with the slurm-backup script

  • An issue that prevented booting over a VLAN

  • A non-fatal error being logged by slurm services due to the deprecated AccountingStorageUser parameter being set in slurm.conf

Removed Features#

  • The Kubernetes Dashboard is no longer supported upstream, and can no longer be installed with cm-kubernetes-setup

Known Issues#

  • Following the update of the cm-openssl package, clusters using cluster-extension may fail to pivot to the local root after an update of the packages in the software image. Administrators should terminate and recreate the compute nodes after the update, or alternatively, add a finalize script for the affected nodes or categories that runs sed ‘s/^dh dh1024.pem/dh none/’ -i /localdisk/etc/openvpn/vpn.0.conf to ensure proper OpenVPN configuration and a successful pivot.

CMDaemon#

New Features#

  • On kubernetes cluster deployment, an unversioned kubernetes module file is generated

Fixed Issues#

  • An issue when cancelling PBS Pro and Open PBS jobs

  • An issue with cmsh when adding categories without arguments

  • An issue with bulk-terminating thousands of compute nodes failing with ResourceCountExceeded when using AWS

  • An issue with the grabimage command that did not grab 10-kubeadm.conf drop-in

  • An issue with the configuration of both hybrid MIG and non-MIG as Slurm resources

cm-setup#

New Features#

  • cm-container-registry-setup sets up Harbor with v2.14

Fixed Issues#

  • An issue where disabling shared public IP creation in cm-cloud-ha-setup was ignored

  • An issue where the OCI signing key was kept in log files