Run Autoscaler with GPU instances on GCP

Hello,
I am trying to get running Drone Autoscaler (DA) on GCP cloud (I am following docs) which would launch agents on the specific instance type. So far I understood that the Autoscaler can run on any type of VM and then it launches its own instances with agents…

  • do I have to create the GCP cloud separately from DA or does DA creates it on its start
  • if DA manages the cloud, how can I specify the instance type I want to run the agent on, I found argument DRONE_GOOGLE_MACHINE_TYPE but this, so far I know, does not specify if GPU is included
  • if above, how can specify the agent instances will be preemptive (even it is not the recommended by Drone due to unexpected termination)

AM am sorry for these trivial questions, but I did not find it in docs nor in the forum…

Hello there,

Kindly let us know on what instance CPU/GPU are you looking to provide as for the GCP there are Machine Types which you can specify once you start the Drone autoscaler on the GCP docs.
Machine Type

Hello, I want to use quite a small CPU/RAM resource-demanding (2 CPU cores and 8GB RAM) with mounted two T4 graphic cards. So the base instance type could be n1-standard-2 but I don’t know how to add GPUs to the specification, somewhere I found also nvidia-tesla-t4 but it does not specify nb or mounted GPUs either…

Update: would the GPU setting be part of https://autoscale.drone.io/configure/cloud-init