Table 3.
IANVS Cluster Specifications.
| Node type | SLURM partition* | Qty. | CPU | Cores (total) | SMT threads (total) | Clock speed (GHz) | RAM (GiB) | Storage | InfiniBand blocking factor** | Remarks |
|---|---|---|---|---|---|---|---|---|---|---|
| login | – | 2 | 2x12-core Intel Xeon E5-2680v3 | 24 | 48 | 2.50 | 256 | GPFS over IB | 1:8 | – |
| small | standard | 180 | 128 | 1:2 or 1:8 | hostnames: small[001-180] | |||||
| large | 76 | 256 | hostnames: large[001-076] | |||||||
| gpu | gpu | 12 | hostnames: gpu[01-12] | |||||||
| special | Special | 2 | 4x 10-core E5-4620v3 | 40 | 80 | 2.00 | 1024 | 1:8 | hostnames: special001, special 002 |
*a “partition” in SLURM terms means “a group of machines to which one may submit cluster jobs”.
**Nodes on the same InfiniBand switch always have their full bandwidth available when communicating with each other. However, if nodes want to communicate over switch boundaries, their available bandwidth might be reduced due to contention on the switch. The “blocking factor” is the maximum reduction of bandwidth that can occur in a case like this.