• cuda-gpu-jon-skarpeteig.png

This graph is from a custom Cuda plugin for CollectD. It collects GPU metrics available through nvidia-smi command line tool. Grafana will dynamically add rows for each GPU core.

The collector can be automatically deployed using Puppet monitoring module

Collector Configuration Details

pip install collectd-cuda

# /etc/collectd/conf.d/python-config.conf
<Plugin "python">
  ModulePath "/usr/local/lib/python2.7/dist-packages"
  LogTraces false
  Interactive false
  Import "collectd_cuda.collectd_plugin"