I have a problem with building TF r2.3 when I would like to build a version with CUDA support. When I configure the build for CPU without CUDA everything build fine. With the CUDA support turned on I got Bad address (Exit 126) error for bash commands like this. I use windows 10 with MSVC 2019 v16.6.5. I also use python 3.6.8 and Bazel 3.3.1. I have encountered with this problem with both CUDA v10.2, cudnn-10.2-windows10-x64-v7.6.5.32 and CUDA 10.1 and cuDNN 7.4. I have tried to build with the following command after I have configured the project.
bazel build --config=opt --define=no_tensorflow_py_deps=true //tensorflow:libtensorflow_cc.so
I put the content of the command.log file to this pastebin link.
I have opened a github issue for this error on the tensorflow repo but so far they haven't provided any solution for this issue. I hope here someone point me in the right direction.
I had the same issue, try to build TF 2.4 with Cuda 11 and cuDNN 8. I was able to build with this configuration on Windows finally.
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With