Keras and TensorFlow programs can be executed on both CPUs and GPUs, though these programs usually run faster on GPUs. If your system does not support an NVIDIA GPU, you only need to install the CPU version. However, if your system has an NVIDIA GPU that meets all the prerequisites and you need to run performance-critical applications, you should install the GPU version. To run the GPU version of TensorFlow, we need an NVIDIA GPU, and then we need to install a variety of software components (CUDA Toolkit v9.0, NVIDIA drivers, and cuDNN v7.0) on the system.

In steps 1 to 3, we created a new conda environment with both the R and Python kernels installed. In steps 4 and 5, we installed the keras library in the environment we created.