Using `-l arch`

As shown by this graph, Hoffman2 Cluster’s compute nodes have various CPU models. Without specifying CPU models, the job scheduler would allocate compute nodes that satisfy the requirements job parameters.

The user can use the option -l arch=... option to specify the CPU model if preferred. For example,

-l arch=intel-gold-*

will force the job to run only on the intel-gold class compute nodes.

Things to consider:

While specifying the CPU model can precisely determine what CPU model(s) the job will be run on, doing so may increase the wait time.
If you specify a CPU model that is not available to your account, the job will not start.
The CPU model needs to be compatible with other job parameters (e.g. memory size, number of cores, etc.)

The CPU models display in the qhost may be truncated. Use qconf -se $hostname to see the full name, e.g.

$ qhost -h n7676
HOSTNAME                ARCH         NCPU NSOC NCOR NTHR NLOAD  MEMTOT  MEMUSE  SWAPTO  SWAPUS
----------------------------------------------------------------------------------------------
global                  -               -    -    -    -     -       -       -       -       -
n7676                   intel-gold-61  36    2   36   36  0.89  188.6G   26.9G    1.9G   80.2M

$ qconf -se n7676 | grep complex
complex_values        arch=intel-gold-6140,exclusive=true,h_data=192G, \

Hoffman2 Cluster Advanced Topics documentation

Using -l arch

Using `-l arch`