Compare GKE Autopilot per-pod billing with Standard node billing, including the cluster fee and the 110-pod ceiling.
Standard mode bills whole nodes whether or not pods fill them, so the node count is the maximum of three constraints: CPU requests, memory requests and the 110-pod ceiling. Autopilot instead bills the sum of your pod requests at a higher per-vCPU rate, which comes out cheaper precisely when your nodes would have sat half empty. The Autopilot-versus-Standard decision usually turns on bin-packing efficiency rather than on the headline rates, and the pod ceiling catches teams running many tiny sidecars long before CPU does. Check node, Autopilot and cluster-fee prices for your region in the Google Cloud pricing calculator before choosing a mode.
Google Kubernetes Engine (GKE)
nodes = max(CPU requests ÷ allocatable CPU, memory requests ÷ allocatable memory, pods ÷ pods-per-node); Standard = nodes × node price × hours + cluster fee, Autopilot = (Σ pod CPU × vCPU rate + Σ pod memory × GiB rate) × hours + cluster fee.
nodes = max(CPU requests ÷ allocatable CPU, memory requests ÷ allocatable memory, pods ÷ pods-per-node); Standard = nodes × node price × hours + cluster fee, Autopilot = (Σ pod CPU × vCPU rate + Σ pod memory × GiB rate) × hours + cluster fee. Standard mode bills whole nodes whether or not pods fill them, so the node count is the maximum of three constraints: CPU requests, memory requests and the 110-pod ceiling. Autopilot instead bills the sum of your pod requests at a higher per-vCPU rate, which comes out cheaper precisely when your nodes would have sat half empty.
The Autopilot-versus-Standard decision usually turns on bin-packing efficiency rather than on the headline rates, and the pod ceiling catches teams running many tiny sidecars long before CPU does. Check node, Autopilot and cluster-fee prices for your region in the Google Cloud pricing calculator before choosing a mode.
This calculator takes 13 inputs: Pods to run, CPU request per pod, Memory request per pod, vCPUs per node, Memory per node, Price per node-hour, Maximum pods per node, Node capacity reserved for the system, Autopilot price per vCPU-hour, Autopilot price per GiB-hour, Cluster management fee, Apply the free cluster credit, Hours in the billing month. The pre-filled defaults are a realistic starting point — replace them with figures from your own environment for a result you can act on.
kubelet, the container runtime, the OS and GKE add-ons such as the metrics agent and DNS all consume resources before your pods are scheduled. The reservation grows with node size in absolute terms but shrinks as a percentage, which is one argument for fewer, larger nodes.