There is now quite a variety of nodes on the cluster. We have both Intel and AMD core nodes, high memory nodes, as well has older and new generation hardware. Depending on the properties of your job you may need to specify these. Note that requests for large amounts of memory should normally be handled through a

#SBATCH --mem #G

or

#SBATCH --mem-per-cpu #G

statement in your sbatch file. Substitute the number of GB you need for “#”. The desired property for nodes can be requested in job submission by including a line like this in your sbatch file.

#SBATCH -C constraint

where the value for “constraint” is taken from the table below. You can specify several constraints at once with AND (&) or OR (|). So, to allocate a new Intel node you could use

#SBATCH -C m630|m640|C6420

See “man sbatch” for more details on specifying this. See /etc/slurm/node.conf for all current nodes and features. We recommend that if the job is using the Intel compiler suite, the “intel” constraint should be specified.

Constraint property preemptable by owners
m610 nodes with x5650 Intel cores no
m620 nodes with e5-2650 or e5-2640 v2 Intel cores yes (on most but not all)
m630 nodes with e5-2640 v3 or v4 Intel cores yes
m640 nodes with 8160 or 6126 Intel cores yes
C6420 nodes with 6126 Intel cores no
intel nodes with Intel cores yes (on most but not all)
amd nodes with AMD cores yes
x5650 nodes with older x5650 Intel cores no
e5-26500 nodes with e5-2650 Intel cores no
e5-2640v2 nodes with e5-2640v2 Intel cores yes
e5-2640v3 nodes with e5-2640v3 Intel cores yes
e5-2640v4 nodes with e5-2640v4 Intel cores yes
8160 nodes with 8160 Intel cores yes
6126 nodes with 6126 Intel cores no (on most but not all)
R920 large memory node with Intel cores yes
R815 large memory node with AMD cores yes
R715 AMD cores yes
R815 AMD cores yes
avx nodes with AVX (Advanced Vector Extensions) support mixed
avx2 nodes with AVX2 support mixed