dnddnjs / pytorch-multigpu

Multi GPU Training Code for Deep Learning with PyTorch

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

pytorch-multigpu

Multi GPU Training Code for Deep Learning with PyTorch. Train PyramidNet for CIFAR10 classification task. This code is for comparing several ways of multi-GPU training.

Requirement

  • Python 3
  • PyTorch 1.0.0+
  • TorchVision
  • TensorboardX

Usage

single gpu

cd single_gpu
python train.py 

DataParallel

cd data_parallel
python train.py --gpu_devices 0 1 2 3 --batch_size 768

DistributedDataParallel

cd dist_parallel
python train.py --gpu_device 0 1 2 3 --batch_size 768

Performance

single gpu

  • batch size: 240
  • batch time: 6s
  • training time: 22 min
  • gpu util: 99 %
  • gpu memory: 10 G

DataParallel(4 k80)

  • batch size: 768
  • batch time: 5s
  • training time: 5 min
  • gpu util: 99 %
  • gpu memory: 10 G * 4

About

Multi GPU Training Code for Deep Learning with PyTorch

License:MIT License


Languages

Language:Python 100.0%