Tensorflow

Tensorflow, utilizing GPU or CPU computation, is available on Koko. Tensorflow is only compatible with version 7 of the Enterprise Linux OS, when selecting partitions, please remember this as the module is not available with the older Enterprise Linux 6 OS. EL7 partitions are denoted with a 7 at the end of the partition name. Partition information can be found here. Please use these instructions for allocating resources:

GPU

  • Allocate resources with the command salloc -p <partition> --gres=gpu
    • The gres:gpu number indicates how many GPUs you’d like to allocate, the default is 1.
  • Find the node allocated using the command ‘squeue’ by your name, it will normally be close to the bottom for newly allocated resources.
  • SSH to your allocated node with the command ‘ssh <node> -X’.
  • Load the modules required with the command ‘module load shared tensorflow-gpu/0.12’
  • Start Python.

CPU

  • Allocate resources with the command salloc -p <partition>
    • Partition information can be found here.
  • Find the node allocated using the command ‘squeue’ by your name, it will normally be close to the bottom for newly allocated resources.
  • SSH to your allocated node with the command ‘ssh <node> -X’.
  • Load the modules required with the command ‘module load shared tensorflow/0.12’
  • Start Python.
Posted in How-To-Guides, HPC