AWS has hundreds of instance families based on different types of CPU architectures. While there are many options with CPU architectures that are optimised for specific workloads (like ML etc), there are three major categories in which all general purpose processors fall. Intel, AMD and Graviton.

What’s the difference?

Unless you are working on something where x86 architecture will really boost the performance, try running your workload first on Graviton.

Intel: Is the the most widely used processor on the planet. Intel was the only choice on AWS until 2018. It is based on x86 architecture and offers solid general purpose performance. These are expensive than AMD and Graviton chips.

AMD: AWS partnered with AMD in 2018 and introduced their first AMD processor as m5a family of instances. AMD and Intel are both based on x86 and therefore offer similar compatibility. AMD however offers better performance than their Intel counterparts and is roughly 10% cheaper than comparable Intel processor on AWS.

Graviton: Custom chip designed by Amazon, based on ARM architecture. For same(or better) performance, they are roughly 20% cheaper than Intel processors. While software compatibility is not as wide as Intel, it is still pretty good. You should choose Graviton if your workload can run on it. They come with larger L1 and L2 cache meaning better performance. Every vCPU is a physical core leading to more consistent performance.

Type Architecture Compatibility with software libraries Performance Cost
Intel x86 Very Good Good general performance Most expensive
AMD x86 Very Good Especially good for multithreaded applications. Roughly 10% cheaper than Intel
Graviton ARM Decent, getting better Best performance for most workloads. Roughly 20% cheaper than Intel. Extremely power efficient.

👉 Some Tips to Save Costs

  1. Plan to migrate to a common instance family slowly over next 6 months or 1 year. Align this timeline with your savings plan renewal dates to get the maximum benefit and avoid undesired lock in. See Standardising Instance Families to Reduce Costs
  2. If your workload can run on Graviton processors, it is better(in most cases) to go with them. They were launched first in 2018 and have been battle tested since then. You’ll get roughly 20% savings without doing any code changes or committing any usage to AWS.
  3. EC2 instances savings plan bought for m6g.large instance will not apply to m6a.large as these are different families.

Next to Read

Data Transfer Costs - Overview

🌼🌼🌼🌼🌼🌼🌼🌼🌼🌼