can you build the tensorflow 1.8 with cuda 9.2? and publish the wheel

[gaohongfein]

Hi,
thanks for you works build the tensorflow with newest cuda.
can you build the tensorflow 1.8 with cuda 9.2? and publish the wheel

[DzianisH]

+1

[gaohongfein]

i have build with cuda 9.2

[sudodaksh]

I'm building, will post the results if i have any.

[alejandrohall]

Hi @DakshMiglani

Did you compile it? Thanks in advance!

[sudodaksh]

@alejandrohall didn't work sorry.

[beew]

I have successfully built and run tensorflow1.8 and 1.9rc1 against cuda9.2+patch1 and cudnn 7.1, with python3.5 on Ubuntu 16.04. I installed cuda9.2 stuffs in a separate test folder (use the .run files without sudo)

I source this script when building and whenever I run these versions of tf (tf1.9 compiled with openmpi. but need to change line 76 of tensorflow/tensorflow/contrib/mpi_collectives/kernels/mpi_ops.cc from se to stream_executor or build would fail)

$PYTHONUSERBASE is set to the test folder so pip3 install --user would install the test tf whl (only one of 1.8 or 1.9rc can exist of course)inside the test folder without messing up the system's version. To invoke it would need to prepend $PYTHONPATH accordingly.

This way it would invoke the test version of tf and it would point to the matching version of cuda (9.2 instead of system's 9.1)

export PREFIX=/home/beew/opt/cuda_test/cuda92
export PATH=$PREFIX/cuda/bin:$PREFIX/bin:$PATH
export CUDA_SDK_ROOT_DIR=$PREFIX/samples/common
export TENSORRT_PATH=$PREFIX/TensorRT-4.0.1.6

export LD_LIBRARY_PATH=$PREFIX/cuda/lib64:$PREFIX/cuda/extras/CUPTI/lib64:$LD_LIBRRAY_PATH:$TENSORRT_PATH/lib

export PYTHONUSERBASE=$PREFIX

export PYTHONPATH=$PREFIX/lib/python3.5/site-packages:$PYTHONPATH

export MPI_HOME=/usr/lib/openmpi

export CPATH=$PREFIX/include:$CPATH
export LIBRARY_PATH=$PREFIX/lib:$LIBRARY_PATH
export LD_LIBRARY_PATH=$PREFIX/lib:$LD_LIBRARY_PATH

alias nvblas92="LD_PRELOAD=$PREFIX/cuda/lib64/libnvblas.so"

It seems that 1.8 without mkl is faster than 1.8 with mkl same phenomenon with 1.9rc1

But then I have only one gpu, maybe the multiple gpu stuffs don't work?

[missionfission]

I have built a Tensorflow-1.9 with Cuda-9.2 here https://github.com/missionfission/tensorflow-wheel

[zychen423]

Well, I found error when installed using @missionfission's build.
So I build it by myself, TF 1.9 with CUDA 9.2
https://github.com/chen0423/TF-1.9-cp36-cuda9.2-wheel

Hope this will help anyone.

[pranman]

@chen0423 This just saved me a few hours of compile time, cheers 👍

[lan2720]

@chen0423 Can you provide more information about how to build by self? Thank you. When I used your .whl, an error raise "tensorflow-1.9.0-cp36-cp36m-linux_x86_64.whl is not a supported wheel on this platform.".

[zychen423]

@lan2720
The detail of building it may need some time to write. I will try to write it down in couples of week.
But for quick advice for you, I think you can check the version with your pip, python, and OS, making what does your pip support
I might be able to help you if you post these information on.

Moreover, you can take a look at:

1. https://stackoverflow.com/questions/28568070/filename-whl-is-not-supported-wheel-on-this-platform
2. https://stackoverflow.com/questions/38866758/filename-whl-is-not-a-supported-wheel-on-this-platform
3. (edit2) https://www.python36.com/how-to-install-tensorflow-gpu-with-cuda-9-2-for-python-on-ubuntu/

(edit1) In the end, I am thinking about if it's inappropriate to discuss it in this thread, since it's actually issue related to my repo.
If it's the case, then I am sorry about that.

Hope this helps and please forgive my poor language.