Slurm is a Job Scheduler software responsible for organizing the execution order of jobs within the system. The fundamental principle of Slurm operation is that users must submit a Job Script through the Login Node to the Slurm queue. When the job's turn arrives, Slurm dispatches the job to be executed on the Compute Node according to the Partition specified in the Job Script file. Upon completion of processing, the results are stored in the directory from which the job was launched.
Partition A scheme for resource allocation or a grouping of compute nodes designated for different types of job execution services.
Queue The sequential order in which submitted jobs await execution.
| Partition | Node | CPU | GPU | Time Limit |
|---|---|---|---|---|
| CPU | 2 | 192 | - | 168 hours |
| GPU | 2 | 64 | 16 | 168 hours |
| Mixed | 1 | 128 | 8 | 24 hours |
| GPU-H100 | 1 | 128 | 4 | 120 hours |
| Partition | Max CPU/Job | Running Job/User | Submit Job/User |
|---|---|---|---|
| CPU | 32 | 2 | 3 |
| GPU | 4 | 2 | 3 |
| Mixed | 64 | 1 | 2 |
| GPU-H100 | 64 | 1 | 2 |