Resource Queue

ICARUS provides Grace CPUs and GH200 GPUs through the gh200 partition. Jobs share the available resources and may queue according to availability and fair-share priority.

See Compute Tiers for per-job node and time limits. Your approved allocation limits apply even if Slurm displays broader partition limits.

Inspect Resources — Read-Only

sinfo -s -p gh200
sinfo -N -p gh200 -o '%N %P %t %c %m %G'

Example output (excerpt; not a live availability report):

NODELIST        PARTITION STATE CPUS MEMORY GRES
CLUSTER-NODE-01  gh200*    idle  144  960000 gpu:nvidia_gh200_144g_hbm3e:2(S:0-1)
CLUSTER-NODE-02  gh200*    mix   144  960000 gpu:nvidia_gh200_144g_hbm3e:2(S:0-1)

idle means unused, mix means partially allocated, and alloc means allocated. * marks the default partition. Memory is scheduler-configured memory in MiB; GPU resources appear under GRES.

Inspect partition limits or a node in more detail:

scontrol show partition gh200
scontrol show node CLUSTER-NODE-01