Build Tensorflow from source

RTX2080 + UBUNTU19.04 + DRIVER418.56 + CUDA10.1 + CUDNN7.6.0 + TENSORFLOW1.13

1. install driver
2. Install cuda10
3. Install cudnn
   1,2,3 see https://blog.youkuaiyun.com/iamsuperman2/article/details/81605707
4. Install TENSORFLOW, build from source
   https://www.tensorflow.org/install/source

When configure build (.configure), do not select XLA JIT, OpenCL SYCL, and ROCm.

The build may take several hours.

5. Test installation

import tensorflow as tf
tf.test.is_gpu_available(
    cuda_only=False,
    min_cuda_compute_capability=None
)

or

https://stackoverflow.com/questions/38009682/how-to-tell-if-tensorflow-is-using-gpu-acceleration-from-inside-python-shell

If get errors like “ImportError: libcudnn.so.7: cannot open shared object file: No such file or directory”, try

sudo ldconfig /usr/local/cuda/lib64