Can I use tensorflow on Windows10 with c++ and gpu support

[xiaoguai0992]

I wish to use tensorflow on windows10 with c++ and gpu support. Is there any pre build SDK for windows? Or I need to compile from sources?

when I tried to compile on windows with cmake and MSVC, smaple trainer cannot build correctly because of some header files missing, such as "graph.pb.h". Is there any solutions?

ENV:
windows10 vs2015 cmake3.6.2

[gunan]

This information is available in our website.
Please make sure you search for solutions before filing issues:
https://www.tensorflow.org/install/install_windows

[sohaiblaraba]

@gunan Following the link you shared it shows only how to install Tensorflow for Python. However, the question of @xiaoguai0992 was about installing Tensorflow on Windows 10 with C++, where on the website, nothing was mentionned regarding this issue.
Tensorflow is an excellent framework and no body can say the opposit. However, it is mainly focused on python + Linux/Mac. I am a windows user, I use a lot Tensorflow/Keras with python but I'd love to see a demo on how to install Tensorflow with C++ on WINDOWS properly. I know it may not be a priority right now, but a lot of people are learning tensorflow, making their own models, but for deployment, we may want to get the best of it, and c/c++ are better than python (I guess ^^). We may also have a lot of projects using c++ and we want to use tensorflow in these projects (I do).

[gunan]

Sorry for the misunderstanding.
Then, our instructions would be here:
https://www.tensorflow.org/install/install_c

I see that these do not mention windows, but I think we also build c libraries for windows.
@asimshankar how different are the instructions for windows?

[asimshankar]

We do build C libraries for Windows, but only with CPU support. Building GPU-enabled binaries as part of the release is on my TODO list but I haven't gotten around to it yet.

• For CPU, you can download the .dll from: https://storage.googleapis.com/tensorflow/libtensorflow/libtensorflow-cpu-windows-x86_64-1.4.0.zip

• We use bazel to build this, but the environment is a bit complicated. The script we use (which is what is run by the nightly builds as well) is here

So, that sort of summarizes the state we're in with regards to the C API and Windows (and I believe other language bindings like those for C# are using these releases)
I'm happy to provide some pointers to next steps, but we (TensorFlow maintainers) are probably not going to have the bandwidth to do this ourselves in the near future.

• If you really need the C++ API (as opposed to the smaller C API), then a similar process should work, except that you'll use bazel to build the //tensorflow:libtensorflow_cc.so target instead of //tensorflow:libtensorflow.so target.

• For GPU support: The script outlined about should work, as long as the build was configured to use CUDA. But I have not tried that yet.

[tensorflowbutler]

It has been 14 days with no activity and the awaiting tensorflower label was assigned. Please update the label and/or status accordingly.

[nikhilk]

+1 to getting this to be officially listed in the documentation along with GPUs of course. :)

[tensorflowbutler]

A member of the TensorFlow organization has replied after the stat:awaiting tensorflower label was applied.

[jackyko1991]

I just have a look at https://github.com/tensorflow/tensorflow/blob/master/tensorflow/contrib/cmake/tf_shared_lib.cmake

this file seems use a very tricky way to generate dll using cmake.

If you directly follow the instruction in https://github.com/tensorflow/tensorflow/blob/master/tensorflow/contrib/cmake/README.md

The tensorflow_BUILD_SHARED_LIB option is default to be off, I am testing on building shared lib with this option on. cmake gui is recommended under windows enviornment to correctly set the dependencies.

Also this cmake file can generate TensorflowConfig.cmake after INSTALL build, location will be $INSTALL_DIR/lib/cmake, not sure if everything is alright, but worth a test