Optimizing EKS Costs with AWS Graviton

This introduces adopting AWS Graviton (ARM64) instances in Amazon EKS as a structural solution for cloud cost optimization. Graviton delivers up to 40% better price-performance compared to equivalent x86 instances, and is also up to 60% more energy efficient. Graviton generation 5 has now been released, featuring a 1 vCPU = 1 physical core mapping with no resource contention between threads, and high memory bandwidth with DDR5-8800 (Graviton5).

EKS costs consist of the control plane (a fixed ~$73/month per cluster), compute, networking, and observability, with compute costs accounting for the majority. Simply switching the instance type of node groups to the ARM64 family can significantly lower compute costs.

To do this, preparing Multi-Architecture container images is a prerequisite. By configuring a manifest list that includes both amd64 and arm64 images under a single image tag, each node automatically selects the image matching its own architecture.

There are three ways to build multi-architecture images:

• Emulation: Uses QEMU + Docker buildx, requires no additional infrastructure, but build speed can drop by up to 10x
• Native build: Builds directly on ARM infrastructure, fastest and most stable, but requires ARM CI infrastructure
• Cross-compilation: Enables fast builds in supported languages such as Go and Rust, with no separate infrastructure needed