Partitions and resources
This page lists the Taihang I cluster's login, CPU, and GPU partitions and common status commands. Choose resources that match your workload.
Naming
Names normally combine q (queue), cpu or gpu (resource type), and a hardware identifier such as 18i, 23i, a800, l20, 4090, or 5090.
Partition inventory
| Partition | Nodes | Count | Type | Configuration described in the source |
|---|---|---|---|---|
qlogin | login[1-6] | 6 | Login | Login, setup, and short tests |
qfree | vnode[1-4] | 4 | Free CPU | 16 virtual CPU cores and 64 GB RAM per node |
qcpu_18i | bnode[1-30] | 30 | CPU | 2 Intel 4116 CPUs, 24 cores, 64 GB RAM per node |
qcpu_23i | cnode[1-16] | 16 | CPU | 2 Intel 8358 CPUs, 64 cores, 512 GB RAM per node |
qcpu_23if | fnode[1-4] | 4 | Large-memory CPU | 4 Intel 8380H CPUs, 112 cores, 1.5 TB RAM per node |
qcpu_23a | gnode[1-19] | 19 | CPU on GPU nodes | Source lists 2 AMD EPYC 7663 CPUs, up to 48 available CPU cores, 96 total cores, and 512 GB RAM per node; see note below |
qgpu_a800 | gnode[1-4] | 4 | GPU | 8 A800 GPUs, 2 AMD EPYC 7663 CPUs, 112 cores, 1 TB RAM per node |
qgpu_a40 | gnode[5-7] | 3 | GPU | 8 A40 GPUs, 2 AMD EPYC 7763 CPUs, 128 cores, 512 GB RAM per node |
qgpu_3090 | gnode[8-17] | 10 | GPU | 8 RTX 3090 GPUs, 2 AMD EPYC 7763 CPUs, 128 cores, 512 GB RAM per node |
qgpu_4090 | gnode[18-19] | 2 | GPU | 8 RTX 4090 GPUs, 2 AMD EPYC 9654 CPUs, 192 cores, 512 GB RAM per node |
qgpu_l20 | gnode20 | 1 | GPU | 8 L20 GPUs, 48 GB GPU memory each, 512 GB RAM per node |
qgpu_5090 | gnode[21-29] | 9 | GPU | 8 RTX 5090 GPUs, 32 GB GPU memory each, 2 AMD EPYC 9654 CPUs, 192 cores, 1 TB RAM per node |
Inventory clarification
The source's qcpu_23a hardware description is not consistent with the differing GPU-node configurations listed for the same node range. Check the actual node allocation and administrator-provided specifications when selecting resources.
Login nodes
Use login nodes for login, setup, compilation, and short tests. Do not run full calculations or large submission workloads on them. A single process may run for at most 60 minutes.
Time limits
qfree: default job time limit of 3 hours; jobs are terminated at their limit.- Other compute partitions: no default time limit currently specified.
qlogin: maximum 60 minutes per process.
Request a realistic runtime even where no default limit is specified.
Inspect cluster state
The following sinfo listing is an illustrative snapshot, not live availability:
[demo@login2 ~]$ sinfo
PARTITION AVAIL TIMELIMIT NODES STATE NODELIST
qfree* up 3:00:00 4 idle vnode[1-4]
qcpu_23a up infinite 1 drng gnode14
qcpu_23a up infinite 10 mix gnode[4,8-13,15,17-18]
qcpu_23a up infinite 2 idle gnode[16,19]
qcpu_23i up infinite 1 mix cnode10
qcpu_23i up infinite 2 alloc cnode[1-2]
qcpu_23i up infinite 13 idle cnode[3-9,11-16]
qcpu_23if up infinite 1 mix fnode1
qcpu_23if up infinite 1 idle fnode4
qcpu_18i up infinite 1 mix bnode1
qcpu_18i up infinite 28 idle bnode[2-29]
qgpu_a800 up infinite 3 mix gnode[1-2,4]
qgpu_a800 up infinite 1 idle gnode3
qgpu_a40 up infinite 2 idle gnode[6-7]
qgpu_3090 up infinite 1 drng gnode14
qgpu_3090 up infinite 8 mix gnode[8-13,15,17]
qgpu_3090 up infinite 1 idle gnode16
qgpu_4090 up infinite 1 mix gnode18
qgpu_4090 up infinite 1 idle gnode19
qgpu_l20 up infinite 1 idle gnode20
qgpu_5090 up infinite 9 idle gnode[21-29]| Command | Purpose |
|---|---|
sinfo -N | Node-level information |
sinfo -N --states=idle | Idle nodes |
sinfo --partition=PARTITION | A selected partition |
sinfo --help | All options |
Common states: drain means no new jobs are accepted, often for maintenance or faults; alloc means fully allocated; idle means available; down means unavailable; mix means partially allocated.