17418

Optimized Broadcast for Deep Learning Workloads on Dense-GPU InfiniBand Clusters: MPI or NCCL?

Ammar Ahmad Awan, Ching-Hsiang Chu, Hari Subramoni, Dhabaleswar K. Panda
Department of Computer Science and Engineering, The Ohio State University
arXiv:1707.09414 [cs.DC], (28 Jul 2017)

@article{awan2017optimized,

   title={Optimized Broadcast for Deep Learning Workloads on Dense-GPU InfiniBand Clusters: MPI or NCCL?},

   author={Awan, Ammar Ahmad and Chu, Ching-Hsiang and Subramoni, Hari and Panda, Dhabaleswar K.},

   year={2017},

   month={jul},

   archivePrefix={"arXiv"},

   primaryClass={cs.DC}

}

Download Download (PDF)   View View   Source Source   

728

views

Dense Multi-GPU systems have recently gained a lot of attention in the HPC arena. Traditionally, MPI runtimes have been primarily designed for clusters with a large number of nodes. However, with the advent of MPI+CUDA applications and CUDA-Aware MPI runtimes like MVAPICH2 and OpenMPI, it has become important to address efficient communication schemes for such dense Multi-GPU nodes. This coupled with new application workloads brought forward by Deep Learning frameworks like Caffe and Microsoft CNTK pose additional design constraints due to very large message communication of GPU buffers during the training phase. In this context, special-purpose libraries like NVIDIA NCCL have been proposed for GPU-based collective communication on dense GPU systems. In this paper, we propose a pipelined chain (ring) design for the MPI_Bcast collective operation along with an enhanced collective tuning framework in MVAPICH2-GDR that enables efficient intra-/inter-node multi-GPU communication. We present an in-depth performance landscape for the proposed MPI_Bcast schemes along with a comparative analysis of NVIDIA NCCL Broadcast and NCCL-based MPI_Bcast. The proposed designs for MVAPICH2-GDR enable up to 14X and 16.6X improvement, compared to NCCL-based solutions, for intra- and inter-node broadcast latency, respectively. In addition, the proposed designs provide up to 7% improvement over NCCL-based solutions for data parallel training of the VGG network on 128 GPUs using Microsoft CNTK.
Rating: 2.8/5. From 2 votes.
Please wait...

* * *

* * *

Featured events

2018
November
27-30
Hida Takayama, Japan

The Third International Workshop on GPU Computing and AI (GCA), 2018

2018
September
19-21
Nagoya University, Japan

The 5th International Conference on Power and Energy Systems Engineering (CPESE), 2018

2018
September
22-24
MediaCityUK, Salford Quays, Greater Manchester, England

The 10th International Conference on Information Management and Engineering (ICIME), 2018

2018
August
21-23
No. 1037, Luoyu Road, Hongshan District, Wuhan, China

The 4th International Conference on Control Science and Systems Engineering (ICCSSE), 2018

2018
October
29-31
Nanyang Executive Centre in Nanyang Technological University, Singapore

The 2018 International Conference on Cloud Computing and Internet of Things (CCIOT’18), 2018

HGPU group © 2010-2018 hgpu.org

All rights belong to the respective authors

Contact us: