-
Notifications
You must be signed in to change notification settings - Fork 0
Using Bridges2
Command:
Note:
You need to specify the node name, e.g., br011. The account you should use is your PSC account, not your XSEDE account. Also, creating public/private key for PSC is highly recommended to avoid entering your password every time.
We have access to multiple partitions of CPU and GPU compute nodes within the Bridges2 system, which should be used for different purposes.
The lab has its own partition called BatComputer, which has fewer total nodes but is not shared with other groups, and running jobs will not charge the grant allocation.
We also have access to shared PSC partitions GPU, GPU-small, and GPU-shared due to our current grant allocation, which has a limited number of service units (SUs). These have many more nodes and GPUs, but are shared with other research groups, and running jobs on them charges SUs to our grant allocation.
| Partition | Account ID | Node IDs | GPU types | Description |
|---|---|---|---|---|
| BatComputer | bio170034p | dv[001-004] | rtx5000, rtx6000 | Lab cluster |
| GPU | asc170022p | v[001-033] | v100-16, v100-32 | Jobs must request all 8 GPUs in the node |
| GPU-small | asc170022p | v[001-033] | v100-16, v100-32 | |
| GPU-shared | asc170022p | v[001-033] | v100-16, v100-32 | Jobs can request only some GPUs in the node |
You can use the following command to get partition configuration info:
sinfo -o %all -p PARTITION
Example for BatComputer:
interact -A bio170034p -p BatComputer --gres=gpu:rtx5000:1 -N 1 -t 8:00:00 --ntasks-per-node=2
Example for GPU-small:
interact -A asc170022p -p GPU-small --gres=gpu:1 -N 1 -t 08:00:00
-
-A: project id -
-p: partition name -
-t: time (48 hours max limit) -
-N: number of nodes -
--ntasks-per-node: Number of cores to allocate per node [set this value high if you are running into memory issues] -
--gres=gpu:N: number of requested GPUs (replace N with number)
If you would like to request GPUs of a specific type, you can specify --gres=gpu:TYPE:N.