The customer is pointing that the memory max value scale up is based in GB value, while the min memory for scale down is based on GiB.

So, setting both values in GB makes scale down fail:

Skipping ocgc4preplatgt-98fwh-worker-c-sk2sz - minimal limit exceeded for [memory] 

While setting both values in GiB makes the scale up fail.

• API reference says that GBs must be used for memory min/max limits:

https://docs.openshift.com/container-platform/4.12/rest_api/autoscale_apis/clusterautoscaler-autoscaling-openshift-io-v1.html#spec-resourcelimits

• While OCP Cluster Autoscaler documentation points to GiBs:

https://docs.openshift.com/container-platform/4.12/machine_management/applying-autoscaling.html#cluster-autoscaler-cr_applying-autoscaling