Eventually, we are going to switch: from this date onward, RHEL 9 will be the default for all partitions,

This always implicates the switch to the new module system (based on SPACK).

There are no special partitions any longer to target RHEL 9 compute nodes. For all jobs compatible with RHEL 9 to be submitted after the 16^th of October, remove any

#SBATCH --partition special_rhel9…

line from your job scripts.

Plan of the Day

Cluster operations will be halted for this process.

Pending jobs will be moved to new partitions if required, meaning that for instance jobs that wait in a partition with RHEL 8 nodes will administratively be moved to a new special partition with RHEL 8 nodes.
As there will be no MEM nodes with RHEL 8, and not all GPU types will be available for RHEL 8, this might not be possible for a few jobs with these requirements.

Jobs waiting in special RHEL 9 partitions will remain in these partitions, and will be able to start and finish in these partitions, once cluster operations are resumed.

However, no new jobs can then be submitted to these partitions. Instead, jobs for RHEL 9 nodes will use the default partitions after the switch, meaning that they should be submitted without requesting any partition.

Compute Nodes under RHEL 8

Instead, you need to specify a special partition if you still need RHEL 8:

#SBATCH -p special_rhel8

for general (MPI) jobs (ca. 128 compute nodes), and

#SBATCH -p special_nvd_rhel8

for GPU jobs (2 accelerator nodes with NVidia A100 GPUs).

Login Nodes under RHEL 8

The two login nodes remaining on RHEL 8 are:

 lcluster7.hrz.tu-darmstadt.de (LB2A2, 104 CPU cores)
lcluster19.hrz.tu-darmstadt.de (LB2A1,   96 CPU cores)