Having trouble instantiating a GPU instance with correct driver (>= 555) on GCP

I would like to learn more about how to reliably setup a Nvidia L4 GPU with the correct driver (>= 555) on Google Cloud (GCP)? I would like to use such a G2 instance, since it seems to be the only ‘cheap’ option for now. The FAQs mention the L4 as a valid option, but tbh I am having great trouble getting this to work (By default all L4 GPUs come with a earlier Nvidia driver and the update to the latest driver somehow fails or is not supported). Is there a recommended way to do this? Are there better alternatives?
Any kind of help would be greatly appreciated :fire:

L4 GPUs are definitely supported, I use one on AWS regularly to develop against MAX. Odd that the driver update is causing such a problem, with the Ubuntu 22.04 or 24.04 images on AWS, I recall it being pretty straightforward.

In any case, we provide Docker images that host a fully-contained MAX environment and that should work great just to get started with MAX on an L4. In our cloud deployment guide you can choose the GCP tab and it should give you step-by-step instructions for running this Docker image on a GCP instance. We in fact use L4 GPUs by default in that walkthrough.

Let me know if you still run into trouble, and I can bring in one of our GCP experts to see if we can resolve this.

3 Likes