topic You can do the sanity check in Intel® Distribution for Python*

Intel-optimized TensorFlow available in default Anaconda Channel now!

Preethi_V_Intel — Thu, 26 Jul 2018 23:45:26 GMT

Updated 8/9/2018: Intel optimized TensorFlow 1.9 wheels and conda packages in Intel channel are made available now! Refer the Install guide for the installation instructions to get the latest Intel Optimized TensorFlow.

Starting from TensorFlow v1.9, Anaconda will continue to build TensorFlow with MKL-DNN optimizations and distribute in their main channel.to deliver a high performance TensorFlow.

Made currently available for Linux platform, if you are using conda package manager, Just follow the instruction below to get the latest and highly optimized TensorFlow.

conda install tensorflow

conda install tensorflow -c anaconda

If you are not using conda package manager, Intel will distribute the latest version(1.9) of optimized-TensorFlow wheels shortly. However, older version wheels are made available. For more details, Click Here

Hi Preethi,

rajaraman__srinivasa — Tue, 07 Aug 2018 16:03:56 GMT

Hi Preethi,

I installed tensorflow in my Windows 10 through conda install -c anaconda tensorflow. Now how do I make sure that this tensorflow build is using Intel MKL-DNN primitives. In other words, is there a command inside tensorflow to check this similar to checking if tensorflow is using GPU.

Best

s0r2637

Hi Srinivasan,

Preethi_V_Intel — Thu, 09 Aug 2018 23:54:05 GMT

Hi Srinivasan, Right now the optimized TF 1.9 is available only for Linux OS

In a freshly created anaconda

Kloppenburg__Ernst — Fri, 16 Nov 2018 10:17:16 GMT

In a freshly created anaconda installation, in an environment created with

conda create -n tensorflow_env tensorflow keras

I still get the warning "cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA"

conda list tensorflow shows

tensorflow: 1.12.0-mkl_py36h69b6ba0_0

which looks promising.

Other people seem to have the same issue, see discussion at https://www.anaconda.com/blog/developer-blog/tensorflow-in-anaconda/

Are there any issues with the packaging?

Hello,

Kloppenburg__Ernst — Fri, 16 Nov 2018 13:36:39 GMT

Hello,

I guess the tensorflow packages from the main anaconda channel have some problem.

Today I installed them, using a freshly downloaded anaconda distribution on linux, in a fresh environment.

With these packages I get the warning:

I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA

'conda list tensorflow' outputs

tensorflow 1.12.0 mkl_py36h69b6ba0_0

doing 'conda install tensorflow -c intel' kind of upgrades me to this warning message:

I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA

here 'conda list tensorflow' shows:

tensorflow 1.11.0 py36_0 intel

you can ignore the warning,

Preethi_V_Intel — Sat, 05 Jan 2019 01:31:35 GMT

you can ignore the warning, since MKL-DNN will use the latest instruction sets. Please refer the install guide for additional details

Please add build for TF 1.14

Bob — Sun, 02 Jun 2019 10:50:00 GMT

Please add build for TF 1.14 (or at least 1.13) / python 3.7

please add Tensorflow 1.14

Bob — Tue, 04 Jun 2019 17:35:55 GMT

please add Tensorflow 1.14 (or at least 1.13) build for python 3.7

You can do the sanity check

johnson__ashish — Tue, 03 Dec 2019 12:37:51 GMT

You can do the sanity check using the below command as given in the intel developer website:

python -c "import tensorflow; print(tensorflow.pywrap_tensorflow.IsMklEnabled())" Just run this command if the result is true then intel optimized tensorflow is being used.