Chimera Scheduler
.
Slurm
Chimera is using the Slurm scheduler. There is significant documentation available for slurm commands via the man pages. Try 'man sbatch' for submission related commands and 'man squeue' for status queries.
Partitions
Partitions and billing weights are shown below. The 'sinfo' command will show current partition status. The billing weights show the relative 'cost' (as manifest in the impact use has on fairshare priority) of CPU, memory, or GPU for each node.
Partition Name | #Nodes | Billing weights | ||
cpu | mem | gpu | ||
Intel2650 | chimera01 | 4.785 | 0.297G | |
AMD6128 | chimera[02-04] | 0.438 | 0.219G | |
AMD6276 | chimera[06-08] | 0.403 | 0.1G | |
Intel6126 | chimera[09-10] | 28.208 | 0.875G | |
Intel6248 | chimera11 | 25.950 | 1.345G | |
DGXA100 | chimera[12-13] | 112.916 | 7.003G | 1806.649 |
Intel6240 | chimera[14-16] | 18.538 | 1.153G | |
Intel | chimera[09-11,14-16] | 28.208 | 1.355G |
The Intel partition includes all newer Intel nodes, and is billed at the most expensive rate of those nodes. Slurm will not schedule an individual job or task on multiple partitions. This partition is only recommended for MPI jobs where you are ok spanning multiple node types. If you want your job to run on any Intel node, but not span multiple nodes with different CPUs, specify a comma separated list of partitions to be considered.
Healey Library, Lower Level
UMass Boston
100 Morrissey Blvd.
Boston, MA 02125
Book a Consultation
617.287.5399
It-rc@umb.edu