Skip to content

Add cudnn 5.0.5 to supported versions#321

Merged
soumith merged 1 commit intopytorch:masterfrom
szagoruyko:cudnn-5-0-5
Dec 17, 2016
Merged

Add cudnn 5.0.5 to supported versions#321
soumith merged 1 commit intopytorch:masterfrom
szagoruyko:cudnn-5-0-5

Conversation

@szagoruyko
Copy link
Contributor

I ran tests locally, only rnn do not pass because they were added in 5.1, the rest pass. Having 5.0 is very useful for:

  • reproducing lua torch results (eg fb.resnet.torch)
  • as far I understand there is fused winograd which I'd like to stay away from, and there's no way to block it in the bindings right now.

@ngimel
Copy link
Collaborator

ngimel commented Dec 16, 2016

fused winograd (a.k.a. just winograd) is in cudnn v5, v5.1 added nonfused winograd. Are you seeing convergence problems with either of them? Also, are there reasons to specify patch versions for 5.1 or 5.0, won't major/minor be enough?

@soumith
Copy link
Collaborator

soumith commented Dec 16, 2016

is there a way to not use Winograd via benchmark or fastest?
Not the first time i've seen someone worry about winograd implementation that might be an issue.

cant do just 5.1 or 5.0 because the libcudnn.so.x doesn't come only in major versions. there's a .so.5 and a .so.5.0.5

@ngimel
Copy link
Collaborator

ngimel commented Dec 16, 2016

Ah, you are right with respect to numbering, my bad. Nonfused winograd is not returned in fastest (cudnnGet*), and benchmark (cudnnFind*) returns a table of results, so they can be filtered in the framework if someone does not want to use Winograd or non-fused winograd. No way to prevent cudnnGet from returning (fused) winograd, AFAIK.

@soumith soumith closed this Dec 17, 2016
@soumith soumith reopened this Dec 17, 2016
@soumith soumith merged commit 5586f48 into pytorch:master Dec 17, 2016
@soumith
Copy link
Collaborator

soumith commented Dec 17, 2016

thanks!

@szagoruyko
Copy link
Contributor Author

@ngimel right thanks, I meant nonfused Winograd, fused one seems to be fine.
@soumith if we implement it, how should it be handled in user API?

@szagoruyko szagoruyko deleted the cudnn-5-0-5 branch December 18, 2016 23:15
KyleCZH pushed a commit to KyleCZH/pytorch that referenced this pull request Sep 20, 2021
Update upload_binary_sizes.sh to use CUDA 9.2
eellison pushed a commit to eellison/pytorch that referenced this pull request Jun 29, 2022
* Initial implementation of UnspecializedPrimitiveVariable

* Update heuristic

* Add test for no recompilations for different values
hubertlu-tw pushed a commit to hubertlu-tw/pytorch that referenced this pull request Nov 1, 2022
Chao1Han pushed a commit to Chao1Han/pytorch that referenced this pull request Jul 31, 2025
Chao1Han pushed a commit to Chao1Han/pytorch that referenced this pull request Aug 25, 2025
Chao1Han added a commit to Chao1Han/pytorch that referenced this pull request Sep 10, 2025
Chao1Han pushed a commit to zhangxiaoli73/pytorch that referenced this pull request Jan 12, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants