End of Life and Milan Nodes in NEMO1
Published: 15 Nov 2024 by NEMO Team
UPDATE 19.12.2024: No queue needs to be specified to submit jobs to the Milan node/partition. With or without the -q milan
option, the jobs are sent to the same nodes.
The AMD Genoa partition for NEMO2 will arrive on December 4. To make room for the new nodes, all old NEMO1 nodes will be decommissioned.
To ease the transition from NEMO1 to NEMO2, one rack with the new NEMO2 nodes has been temporarily reconfigured to run the NEMO1 environment. As an added benefit, this will give more processing power to NEMO1 for the time being. If we see demand, we can add a total of 17000 cores to NEMO1.
We will keep these AMD Milan nodes in the NEMO1 environment during the migration to NEMO2.
To run jobs on the new nodes, you have to explicitely request them by specifying the milan
queue:
msub -q milan ...
or in the job script
#MOAB -q milan
Because each Milan node has 128 physical cores, they are configured to run in shared mode.
If your job does not require all 128 cores of a Milan node, it is likely that your job is sharing that node with other users’ jobs.
Each node has approximately 500 GB of RAM, which means you can use approximately 3.9 GB per core.
If your jobs do not require more RAM, we recommend that you use -l pmem=3900M
or less.
Of course, you can always use more memory if needed.
More information about the Milan nodes can be found in the old news.