cortex
cortex copied to clipboard
Decide on the resource requests/limits for Neuron device plugin and scheduler
Description
Change to appropriate resource requests and/or limits for Neuron k8s device plugin. The AWS team has said their device plugin uses few CPU/Mem resources, but that's still not a number. To be done in manager/manifests/inferentia.yaml
.
Currently, we've associated 100m
CPU time & 100Mi
memory for the device plugin. Change that to an appropriate value once more information is released by them.
Additional context
As discussed in https://github.com/aws/aws-neuron-sdk/issues/103.