Nowadays, you can rent powerful machines with GPUs by the hour. Pricing varies, depending on the GPU power and the provider. It usually costs around $1 per hour for a single GPU machine, with the price going down every day. If you commit to renting the machine for the month, you can get good computing power for around $100 per month. Considering the time you will save waiting for the model to train, it often makes economic sense to rent a remote machine.
Another option is to build your own deep learning server. Note that this requires investment and assembly, and that GPUs consume large amounts of electricity.
Once you have secured access to a remote machine, you have two options:
- Run Jupyter Notebook on the remote server. Jupyter Lab or Jupyter Notebook will then be accessible using your browser, anywhere on the planet. It is a very convenient way of performing deep learning.
- Sync your local development folder and run your code remotely. Most IDEs have a feature...
