CUDA10 on Ubuntu 18.04

With a previous installation of CUDA9 I wrote code for my GPU against CUDA directly, but mostly using TensorFlow and Keras. To upgrate to CUDA10, I heard that Anaconda was a quicker way to install the dependencies, including cuDNN. It was relatively simple, but required manual installation of the latest libraries - at least on Ubuntu.

First, find the version you want to install and its md5sum. They're listed here, and I'm using the latest (March 2019) values for Linux.

wget https://repo.anaconda.com/archive/Anaconda3-2019.03-Linux-x86_64.sh
echo '43caea3d726779843f130a7fb2d380a2 Anaconda3-2019.03-Linux-x86_64.sh' | md5sum -c
# If hash was OK, install
bash ./Anaconda3-2019.03-Linux-x86_64.sh

Towards the end of the installation, you'll be informed that:

==> For changes to take effect, close and re-open your current shell. <==

If you'd prefer that conda's base environment not be activated on startup,
   set the auto_activate_base parameter to false:

conda config --set auto_activate_base false

Finally, to install the required libraries - including cuda and cuDNN (versions 10.0.130 and 7.3.1 for the above installation):

(base) iain@iain-ubuntu:~$ conda install keras tensorflow-gpu

If you've already installed a previous version of the CUDA libraries, you might have a problem with conflicts.

python -c "from keras import backend as b; b.tensorflow_backend._get_available_gpus()"

Fails with tensorflow.python.framework.errors_impl.InternalError: cudaGetDevice() failed. Status: CUDA driver version is insufficient for CUDA runtime version

The solution is to install the latest drivers from NVIDIA. Once complete, the latest versions installed are reported for me as:

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 418.56       Driver Version: 418.56       CUDA Version: 10.1     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 106...  Off  | 00000000:01:00.0  On |                  N/A |
|  0%   51C    P0    30W / 120W |    314MiB /  6077MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+

And the available devices now include the GPU:

2019-04-17 12:22:41.466674: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1115]
Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 4509 MB memory)
-> physical GPU
(
  device: 0,
  name: GeForce GTX 1060 6GB,
  pci bus id: 0000:01:00.0,
  compute capability: 6.1
)

⁂