Skip to content

[Speed] Refine parallel_do_grad #9072

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 1 commit into from
Mar 15, 2018
Merged

[Speed] Refine parallel_do_grad #9072

merged 1 commit into from
Mar 15, 2018

Conversation

chengduoZH
Copy link
Contributor

@chengduoZH chengduoZH commented Mar 14, 2018 •
edited
Loading

fix #9074

@chengduoZH
Copy link
Contributor Author

  • model: SE-ResNet-150
  • Input: 3 x 224 x 224
  • batch_size: 12
  • Profiling machine
  ncclAllReduce ncclReduce  
2 cards 0.976438038 0.978825315 0.99756108
3 cards 1.136392013 1.074367762 1.057730931
4 cards 1.274742246 1.170658557 1.088910373

the result shows that, as the number of cards increasing, ncclReduce is faster than ncclAllReduce for parallel_do_grad.

tonyyang-svail
tonyyang-svail approved these changes Mar 14, 2018
View reviewed changes
Copy link

@tonyyang-svail tonyyang-svail left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks good to me.

@chengduoZH chengduoZH merged commit 11c43e5 into PaddlePaddle:develop Mar 15, 2018
@chengduoZH chengduoZH changed the title Refine parallel_do_grad [Speed] Refine parallel_do_grad Mar 15, 2018
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@tonyyang-svail tonyyang-svail tonyyang-svail approved these changes

@reyoung reyoung Awaiting requested review from reyoung

@QiJune QiJune Awaiting requested review from QiJune

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

Refine parallel_do_grad

2 participants

@chengduoZH @tonyyang-svail