Note: For the complete documentation, see this list of system partitions.
Unity is split into multiple partitions. In order to use a specific partition, use the following flag in your job submissions:
-p <partition name>
It is also possible to specify a list of partitions. For shortest wait times, and best quality of service, use one of the following:
For CPU jobs lasting up to a day: -p uri-cpu,cpu,cpu-preempt
For CPU jobs lasting more than a day: -p uri-cpu,cpu-long,cpu-preempt
For GPU jobs lasting up to a day: -p uri-gpu,gpu,gpu-preempt
For GPU jobs lasting more than a day: -p uri-gpu,gpu-long,gpu-preempt
Note that if the other queues are busy and your job ends up in the preempt queue, then the job may be terminated after two hours of running if the resources need to be reclaimed for the purchasers of those nodes. (This also happens for non-URI users using our nodes as preempt.)
Partitions
All general partitions except preempt partitions have a hard limit of 1000 CPUs and 64 GPUs per lab . URI members have access to two priority partitions: uri-cpu and uri-gpu that have no such limits.
Note: Interactive jobs are limited to 8 hours. Use Batch jobs for longer work.
| Name | Time Limit | Max CPUs per Node | Comments |
|---|---|---|---|
| uri-cpu | 30 DAYS | 64 | CPU priority partition |
| uri-gpu | 30 DAYS | 64 | GPU priority partition |
| cpu-long | 14 DAYS | 40 | CPU general partition |
| cpu | 1 DAY | 40 | CPU short partition |
| cpu-preempt | 14 DAYS | 256 | Priority jobs can preempt (requeue) your jobs at any point, so use only if your job supports checkpointing |
| gpu-long | 14 DAYS | GPU general partition | |
| gpu | 1 DAY | GPU short partition | |
| gpu-preempt | 14 DAYS | Priority jobs can preempt (requeue) your jobs at any point, so use only if your job supports checkpointing |
There are other partitions that give priority to other groups who have purchased hardware.