Search code examples
google-cloud-platformpytorchgoogle-compute-enginegoogle-dl-platform

using Deep Learning VM without GPU?

I'm trying to learn how to do ML/PyTorch on Google Cloud. The PyTorch docs recommend the Deep Learning VM, but I have a GPU quota of 0, and even if my quota allowed it, I don't really want to spend $300/month just to go through some PyTorch tutorials.

How can I configure this VM for CPU only? Or is there another, similar "click to deploy" PyTorch setup that does not require a GPU?

Solution

  • There are three ways you can accomplish your task:

    1. Use Vertex AI preconfigured VMs Go to Workbench - Enable API - New Instance - User Managed - PyTorch They are cheap and come with all libraries needed. You don't need to add GPUs.

    2. Use Google Cloud Skills Boost You can also pay a monthly fee of 29 USD to have access to many LABs using Vertex AI. But you will have to find PyTorch LABs.

    3. You can go to Compute Engine and create a new VM, changing the image of the VM to a Deep Learning one. You will need to SSH and also install Jupyter.

    For all of them, the cost will depend upon the number of cores you choose. For me, Vertex AI is more adequate, as you already have Jupyter notebooks installed and does not need to SSH, as it has JupyterLab.

    Note: N1 instances are the cheapest.