Trying the install script for Google gcp. Was granted 1 gpu after request.
Now getting
ERROR: (gcloud.compute.instances.create) Could not fetch resource:
The zone ‘projects/firm-tracer-244111/zones/us-west2-b’ does not have enough resources available to fulfill the request. Try a different zone, or try again later.
So some comments about it, but not sure what is the simplest way to solve it.
I have also been having this issue with an existing instance (–zone=us-west2-b) since yesterday afternoon. It’s still not working. Other online discussions suggest that 6 hours is a reasonable time to wait. But this exceeds that.
I have created a similar instance for east4 and that’s working, but it was a very time-consuming process since my notes on setup were inaccessible in the west2 VM.
Has Google been changing its allocation policies? Should fastai setup guide no longer advise everyone to select west2-b?
I’m seeing the same issue with my GCP instance. I found this link to move zones for existing instances but seems like a risky proposition… requires moving persistent storage and reattaching to new instance!
For each GPU, it tells you which zones offer it. Thus, since https://course.fast.ai/start_gcp.html recommends using a P4, you can see that currently the zones equipped with it are
us-west2-c
us-west2-b
us-central1-a
us-central1-c
us-east4-a
us-east4-b
us-east4-c
northamerica-northeast1-a
northamerica-northeast1-b
northamerica-northeast1-c
europe-west4-b
europe-west4-c
australia-southeast1-a
australia-southeast1-b
asia-southeast1-b
asia-southeast1-c
Thus, when us-west2-b is not available (which is happening pretty often lately), you can start trying from us-west2-c and keeping moving further away geographically until you find a zone which has a P4 resource available. That’s what I do, and it works pretty fine for me.
Thanks so much for your reply, @AndreaPi! Mine’s working today, too, but I’m grateful to have these instructions to fall back on if it goes wonky again.