EC2 Instance Type Selection

Choosing the right EC2 instance family for compute, memory, storage, and accelerated workloads.

Overview

What is being compared

EC2 offers over 600 instance types, grouped into families tuned for different workload profiles. Choosing poorly means paying for capacity you do not use, or hitting bottlenecks you cannot escape without downtime.

Balance — CPU and memory balanced — Higher CPU ratio — Built for in-memory data — High IOPS storage — GPU compute
Typical fit — T: burstable dev and low-traffic work. M: web and application servers — CPU-bound batch processing and HPC — In-memory databases and caching — High IOPS storage workloads — Machine learning and GPU compute

Families by bottleneck

Start from the workload's bottleneck. Balanced workloads go to T or M, CPU-bound ones to C, memory-bound ones to R, storage IOPS to I, and GPU work to P or G.

Burstable T instances

T3 and T4g instances earn CPU credits while they run below their baseline and spend them during bursts. What happens when the credits run out depends on the credit mode:

Which one to pick

Check the bottlenecks in order on a profiled workload. GPU work goes to P or G, a storage-latency bottleneck to I, a memory-bound workload to R and a CPU-bound one to C. A workload with genuine idle periods and low traffic fits T; everything else starts on M.

> [!TIP] The six questions that pick a family, the path from profiling to a commitment, and local NVMe next to EBS, step by step in Visual Explanation below.