Tensorflow compiling from source with GPU

gistfile1.txt

Compiling TensorFlow v2.13.0 from source:
This is in conjunction with https://www.tensorflow.org/install/source but fills the gaps!

Install Bazel 5.3.0

Install the cuda toolkit v11.8 and get rid of any other cuda toolkits. Otherwise TF will find them and create chaos.
apt install cuda-toolkit-11-8

Install cudnn 8.7 (TF recommends 8.6 but 8.7 also works well):
apt-get install libcudnn8-dev=8.7.0.84-1+cuda11.8 libcudnn8=8.7.0.84-1+cuda11.8
apt-mark hold libcudnn8-dev=8.7.0.84-1+cuda11.8 libcudnn8=8.7.0.84-1+cuda11.8

apt install patchelf
pip install --upgrade protobuf

lspci | grep VGA
Get the compute capability of the card here: https://developer.nvidia.com/cuda-gpus
for example 6.1

./configure
[ anything default except below ]

Do you wish to build TensorFlow with CUDA support? [y/N]: y

Please specify a list of comma-separated CUDA compute capabilities you want to build with.
[Default is: 3.5,7.0]: 6.1

bazel build --local_ram_resources=2048 //tensorflow/tools/pip_package:build_pip_package

./bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg

sudo pip install /tmp/tensorflow_pkg/tensorflow-2.13.0-cp310-cp310-linux_x86_64.whl