EC2 Instance Type Guide
Decode AWS EC2 instance names. Learn what each letter, number, and size means so you can pick the right instance for your workload.
How Instance Names Work
Instance Family
Defines the hardware category (general purpose, compute, memory, etc.)
Generation
Higher = newer hardware, better price/performance. Always prefer latest gen.
Processor Suffix
Optional. Indicates processor type or special capability (AMD, Graviton, NVMe, etc.)
Size
Determines vCPU and memory. Each step up roughly doubles both.
Instance Families
General Purpose
Balanced compute, memory, and networking. Best for web servers, app servers, small databases.
Compute Optimized
High CPU performance. Best for batch processing, gaming servers, scientific modeling, HPC.
Memory Optimized
High memory-to-CPU ratio. Best for large databases, in-memory caching, real-time analytics.
Memory Optimized (Extreme)
Extreme memory capacity. Best for SAP HANA, large in-memory databases.
Burstable General Purpose
Baseline CPU with burst capability using CPU credits. Best for dev/test, low-traffic apps.
Storage Optimized
High-speed NVMe SSD instance storage. Best for NoSQL databases, data warehousing.
Dense Storage
High-density HDD storage. Best for distributed file systems, Hadoop, data lakes.
GPU — ML Training
NVIDIA GPUs for ML training and HPC. Best for deep learning, computational finance.
GPU — Graphics & Inference
NVIDIA GPUs for graphics and ML inference. Best for video encoding, game streaming.
AWS Inferentia
Custom AWS Inferentia chips for ML inference at scale. Best for NLP, computer vision.
AWS Trainium
Custom AWS Trainium chips for ML training. Best for large language models.
FPGA
Field Programmable Gate Arrays for hardware acceleration. Best for genomics, financial analytics.
Processor & Capability Suffixes
These letters appear after the generation number and indicate the processor type or a special hardware capability.
| Suffix | Meaning |
|---|---|
| a | AMD EPYC processor — typically 10% cheaper than Intel equivalent |
| g | AWS Graviton (ARM) processor — best price/performance ratio, up to 40% cheaper |
| n | Higher network bandwidth — up to 100 Gbps |
| d | Local NVMe SSD instance storage included |
| e | Extra memory or storage capacity |
| z | High sustained all-core turbo frequency (up to 4.5 GHz) |
| b | Block storage optimized — higher EBS bandwidth |
| i | Intel processor (explicitly marked in newer generations) |
Size Tiers
Sizes below are for the m5 family as reference. Other families follow the same doubling pattern but with different base ratios.
| Size | vCPU | Memory |
|---|---|---|
| nano | 1 | 0.5 GiB |
| micro | 1 | 1 GiB |
| small | 1 | 2 GiB |
| medium | 1 | 4 GiB |
| large | 2 | 8 GiB |
| xlarge | 4 | 16 GiB |
| 2xlarge | 8 | 32 GiB |
| 4xlarge | 16 | 64 GiB |
| 8xlarge | 32 | 128 GiB |
| 12xlarge | 48 | 192 GiB |
| 16xlarge | 64 | 256 GiB |
| 24xlarge | 96 | 384 GiB |
| 48xlarge | 192 | 768 GiB |
| metal | varies | varies |
Quick Tips
Always use latest generation
Newer generations offer better performance at the same or lower price. Prefer m7 over m5.
Try Graviton (g suffix)
ARM-based Graviton instances are up to 40% cheaper with comparable or better performance for most workloads.
T-series for variable workloads
If your app has low average CPU but occasional spikes, t3/t4g instances with burst credits are very cost-effective.
Match family to workload
Don't over-provision. A memory-heavy DB needs r-series, not m-series with extra RAM.
Ready to compare prices?
Search any instance type and instantly see on-demand, reserved instances, savings plans, and spot prices side by side — across every AWS region.
Open Calculator