Max pods per node: meaning and cost impact

• On EKS, most pods take an IP address from the node, so by default the cap is interfaces × (addresses per interface minus 1) + 2.
• Without a custom machine image, managed node groups cap it at 110 below 30 virtual CPUs and 250 above.
• On AKS, each node pool sets it, up to 250. The default is 30 to 250, by network plugin.
• To change max pods per node on AKS, add a pool with the new value and move your pods.

An m5.large has 8 GiB of memory and 3 network interfaces with 10 IPv4 addresses each. EKS allows 3 × 9 + 2 = 29 pods.

Say each pod requests 100 MiB, as an illustration: about 2.8 GiB in all. Pod 30 needs another node, with most memory unclaimed.

You pay for each node, however many pods it runs. When slots fill first, paid CPU and memory sit unclaimed.

• A higher cap, or a bigger EKS instance, allows denser bin packing.
• Coral rebalances CPU and memory requests on AKS and EKS to match real use, and drains spare nodes so your autoscaler can remove them.
• The free assessment works out your own figure.