Skip to content

only install mkl-dnn by default with TensorFlow on x86-64 systems#1666

Merged
boegel merged 8 commits intoeasybuilders:developfrom
bear-rsg:tensorflow-power9
Apr 19, 2019
Merged

only install mkl-dnn by default with TensorFlow on x86-64 systems#1666
boegel merged 8 commits intoeasybuilders:developfrom
bear-rsg:tensorflow-power9

Conversation

@edmondac
Copy link
Copy Markdown
Contributor

This is just a part of the TensorFlow on POWER9 story. See easybuilders/easybuild-framework#2655 and easybuilders/easybuild-framework#2654 for example. Other PRs will follow in time.

@boegel boegel added this to the next release (3.9.1) milestone Mar 26, 2019
Comment thread easybuild/easyblocks/t/tensorflow.py
@akesandgren
Copy link
Copy Markdown
Contributor

On x86_64 system we do want mkl-dnn even if cudnn is available. So please do not drop it for all archs.

@edmondac
Copy link
Copy Markdown
Contributor Author

edmondac commented Apr 9, 2019

Sorry - I was confused by the earlier comment. I've now put that whole block in an if get_cpu_architecture() == X86_64 statement... Is that now right?

Comment thread easybuild/easyblocks/t/tensorflow.py Outdated
Comment thread easybuild/easyblocks/t/tensorflow.py Outdated
Comment thread easybuild/easyblocks/t/tensorflow.py Outdated
@edmondac
Copy link
Copy Markdown
Contributor Author

Is this how you'd like it @boegel ?

Comment thread easybuild/easyblocks/t/tensorflow.py Outdated
@boegel boegel changed the title Tensorflow power9 only install mkl-dnn by default with TensorFlow on x86-64 systems Apr 19, 2019
@boegel
Copy link
Copy Markdown
Member

boegel commented Apr 19, 2019

@edmondac Thank you very much for your persistence on this, I think we've found the right way to tackle this now, so merging...

@boegel boegel merged commit 259a5d9 into easybuilders:develop Apr 19, 2019
@edmondac edmondac deleted the tensorflow-power9 branch April 25, 2019 09:13
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants