Kubernetes Pod Capacity Calculator

Formula: Allocatable = capacity − kube-reserved − system-reserved − eviction threshold. Available = (allocatable − DaemonSet requests) × (1 − headroom). Pods = floor(min(available CPU ÷ pod CPU, available memory ÷ pod memory, pod limit − DaemonSet pods)); EKS pod limit = ENIs × (IPs per ENI − 1) + 2.

Worked example:An 8 vCPU / 32 GiB node under the GKE 1.37+ formula reserves 90m CPU and 2,150 MiB plus 100 MiB eviction. After 3 DaemonSet pods (200m, 400 MiB) and 10% headroom, 27 pods of 250m / 512 MiB fit; CPU is the bottleneck and 60 replicas need 3 nodes.

Common mistake or edge case:A mathematical maximum is not a production target. Account separately for topology spread, ephemeral storage, pod CIDR size, and failure headroom (losing one node must not leave pods Pending).

Additional official references:

• GKE node sizing and allocatable resources
• Amazon EKS: how maxPods is determined
• Amazon EC2 general purpose instance network specifications

Read the supporting guide and worked examples.

For a standalone Docker host, use the Docker memory calculator instead.

Frequently asked questions

Why is my actual pod density lower than this calculator's result?

Pod density can also be capped by the node's pod CIDR, ephemeral storage, topology spread rules, and pods that bypass requests. Enter DaemonSets and EKS ENI values to model the most common caps.

How does EKS decide max pods per node?

With the Amazon VPC CNI and no prefix delegation, the default is ENIs × (IPv4 addresses per ENI − 1) + 2. Managed node groups without a custom AMI cap that at 110 for instances under 30 vCPUs and 250 above. The detail table shows examples.

Which reservation model should I choose?

Use a GKE formula for GKE node pools, a custom value matching your kubelet --kube-reserved and --system-reserved flags elsewhere, or the allocatable option if you copy values from kubectl describe node.