Turi Create does not require a GPU, but certain models can be accelerated by the use of a GPU.
To enable GPU support in Linux after installation of the turicreate package, first make sure both
CUDA and cuDNN are installed.
Make sure to add the CUDA library path to your LD_LIBRARY_PATH environment
variable. In the typical case, this means adding the following line to your
~/.bashrc file:
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATHIf you installed the cuDNN files into a separate directory, make sure to separately add those as well.
The next step is to uninstall tensorflow and install the
CUDA-enabled tensorflow-gpu package:
(venv) pip uninstall -y tensorflow
(venv) pip install tensorflow-gpu
If you have trouble setting up the GPU, the TensorFlow installation instructions may offer additional help.