Running CUDA Scripts on Penn State's ICS-ACI Cluster

Published: by

NOTE: This post is out of date! Penn State has switched to a new cluster system.

This is for my own reference, because if I go more than a few days without doing these steps I will probably forget them.

  1. If there are files on your local computer that you need to use on the cluster, transfer them over with this command.
scp path/to/file/file1.c other/path/file2.c

To copy an entire directory over:

scp -r path/to/directory
  1. The Lion-GA cluster is the one with the Nvidia graphics cards. SSH into it with -Y so that you can launch a text editor with a GUI. I refuse to learn vim or emacs.
ssh -Y
  1. Once you’re in Lion-GA, to see all available GPU nodes, do pbsnodes. There are 8 total as of right now. It seems like there are always a few that are offline. To see which are currently offline, do pbsnodes -l.
  2. SSH into one of the nodes that are online.
ssh lionga1
  1. To see a list of the 8 graphics cards and any processes that are currently running on them, do nvidia-smi.
  2. Once you are SSH’d into the GPU node, you can compile CUDA .cu files using nvcc and run the output, just as you would on your local machine.
  3. To edit a file, logout of the GPU node and then run
gedit file.c