Any plan to provide Intel-optimized packages for TensorFlow 1.0.1?

Any plan to provide Intel-optimized packages for TensorFlow 1.0.1?

When installing Intel Distribution for Python and TensorFlow 1.0.1, I still see TensorFlow is not even optimized for AVX2.

$ python -V
Python 2.7.12 :: Intel Corporation
$ python 
Python 2.7.12 |Intel Corporation| (default, Feb 12 2017, 04:02:31) 
[GCC 4.8.2 20140120 (Red Hat 4.8.2-15)] on linux2
Type "help", "copyright", "credits" or "license" for more information.
Intel(R) Distribution for Python is brought to you by Intel Corporation.
Please check out: https://software.intel.com/en-us/python-distribution
>>> import tensorflow as tf
>>> s = tf.Session()
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE3 instructions, but these are available on your machine and could speed up CPU computations.
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.1 instructions, but these are available on your machine and could speed up CPU computations.
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.2 instructions, but these are available on your machine and could speed up CPU computations.
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations.
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations.
W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use FMA instructions, but these are available on your machine and could speed up CPU computations.
>>>
$

Is there any plan to provide AVX2 package for TensforFlow? Compiling TensorFlow is not a trivial process, that's why providing such package will reach a broader audience.

Thanks!

Manel

Zone: 

Thread Topic: 

Question
10 posts / 0 new
Last post
For more complete information about compiler optimizations, see our Optimization Notice.

Hi Manel,

The version you are using now is most likely the conda forge package, which downloads the binary wheel that google has published.

We are experimenting now with various builds to find the simplest way to get best performance on machines with AVX2 and AVX512. Intel has also submitted some pull requests to the tensorflow project for optimization. We are hoping all issues will be resolved in a few weeks and then we can provide instructions on how to get an optimized tensorflow.

Robert

 

That would be great. Thanks for the quick update!

Manel

Hi Robert,  when there's the Intel optimized Tensorflow available, where can I receive notices on how to download?

thanks

Someone will put a notice here, and that should trigger an email to you.

Hi Manel,

There is going to be a TensorFlow on Modern Intel® Architectures Webinar in May 9, 2017. The following link provides detailed information about the TensorFlow* on Modern Intel® Architectures Webinar

 

Gastón C. Hillar

Manel,

The speaker for this webinar is ElMoustapha Ould-Ahmed-Vall, Senior Principal Engineer in Artificial Intelligence Products Group (AIPG)

Gastón C. Hillar

Quote:

Robert C. (Intel) wrote:

Hi Manel,

The version you are using now is most likely the conda forge package, which downloads the binary wheel that google has published.

We are experimenting now with various builds to find the simplest way to get best performance on machines with AVX2 and AVX512. Intel has also submitted some pull requests to the tensorflow project for optimization. We are hoping all issues will be resolved in a few weeks and then we can provide instructions on how to get an optimized tensorflow.

Robert

 

Hi Robert,

I was getting same warnings reported by Manuel at the beginig of this threat.

Just for fun, I built from sources a CPU tensorflow passing to the configure script the compiler option -march=core-avx2 and to use mkl /home/intel2017/mkl.

Now the warnings of not having support for the instrucctions SSE3, ..., AVX, AVX2 are gone.

How can I check if this build is really using mkl?

Sergio

Enhance your #MachineLearning and #BigData skills via #Python #SciPy
1) https://www.packtpub.com/big-data-and-business-intelligence/numerical-an...
2) https://www.packtpub.com/big-data-and-business-intelligence/learning-sci...

 

All the optimization changes have been merged upstream and you can now build an intel optimized version tensorflow by following the directions here:

https://software.intel.com/en-us/articles/tensorflow-optimizations-on-modern-intel-architecturehttps://software.intel.com/en-us/articles/tensorflow-optimizations-on-modern-intel-architecture

We are working on making a binary package that we can publish, but it should not be difficult to build the package following the directions above on a linux system.

Robert,

Awesome news! I might try by myself whenever I have time, but I will definitely look forward for the public binary package.

Best,

Manel

Leave a Comment

Please sign in to add a comment. Not a member? Join today