Page tree
Skip to end of metadata
Go to start of metadata

At the time of this document, RC has three models of nVidia GPU in the compute cluster:

  • gt740, 2Gb 
  • k20, 4.8Gb
  • Titan, 6Gb

These GPUs are scheduled using the Generic REsource Scheduler (gres) feature of slurm.  This ensures that the only jobs that can talk to a GPU are the ones scheduled against it.  We made a design decision to put all GPUs in a single gres called "gpu" even though we have two basic sizes.  If you simply request a gpu gres, you may get a small or large card depending on availability. 

We leverage a second feature of slurm called constraints to restrict which nodes are elegible to service a job request based on features we assign to specific nodes.  Some of the current features include: rhel, rhel6, rhel7, intel, amd, cuda, bigcuda, infiniband.

The two features we are interested in for GPU scheduling are cuda and bigcuda.  Nodes with small GPUs (gt740) have a feature of cuda. Nodes with big GPUs (k20, titan) have a feature of bigcuda.  This way, you can constrain your job to an approprately sized GPU.


Run the job file against any gpu available in the cluster

sbatch --qos=work --gres=gpu

Run the job file against large memory GPUs

sbatch --qos=work --gres=gpu --constraint=bigcuda