Coding your model on your computer is often the fastest way to get started. As you have access to a familiar environment, you can easily change your code as often as needed. However, personal computers, especially laptops, lack the computing power to train a computer vision model. Training on a GPU may be between 10 and 100 times faster than using a CPU. This is why it is recommended to use a GPU.
Even if your computer has a GPU, only very specific models can run TensorFlow. Your GPU must be compatible with CUDA, NVIDIA's computing library. At the time of writing, the latest version of TensorFlow requires a CUDA compute capability of 3.5 or higher.
Some laptops are compatible with external GPU enclosures, but this defeats the purpose of a portable computer. Instead, a practical way is to run your model on a remote computer that has a GPU.
