@inproceedings{
  author={ A. Awan, C. Chu, H. Subramoni, D. Panda },
  title={ Optimized Broadcast for Deep Learning Workloads on Dense-GPU InfiniBand Clusters: MPI or NCCL? },
  conference={ The EuroMPI 2018 Conference },
  year={ 2018 },
  month={ September },
  location={ Barcelona, Spain },
  source={ http://nowlab.cse.ohio-state.edu/publications/ },
}