Using GPU Nodes
Interactive Mode via login nodes
You can access the GPU login nodes interactively via SSH. First, log into one of our (graphical) cluster frontends and then use SSH to log into one of the interactive (login) GPU nodes.
GPU login nodes are listed here.
Friendly Usage and fair use
The GPU login nodes assign the available GPUs exclusively to processes ("exclusive process" compute mode). Whenever a GPU program is run by any user in the login node, the process gets the whole GPU and does not have to compete with other programs for resources (e.g. GPU memory). Furthermore, several threads in a single process can use the GPUs of a node (cf. e.g. cudaSetDevice) instead of being restricted to one thread per device. Therefore you should use them reasonably. Please run long computations in batch mode using slurm jobs and close any debuggers or Matlab scripts after using the GPU login nodes. You may have a strict runtime limitation on GPU login nodes.
GPU + MPI
If you would like to test your GPU + MPI program interactively, you can do so on the GPU login nodes
C. To get your MPI program to run on the GPU machines, you have to explicitly specify their hostnames. If you are on a GPU login node you can use the
$HOST environment variable to stay on the same node.
Example usage, from a login node equipped with GPGPUs:
$MPIEXEC -np 2 -H $HOSTNAME myApp.exe # 2 processes on the actual node
However, you should not use the '-H' option with regular MPI applications and/or front ends!