ec2 names — a visual guide

Overview

The name tells you what is in the machine.

Renting a server on AWS means picking a name like m7i.large. It looks like a licence plate. It is actually a description, and there are only two things you need to read it.

01

The dot splits it in two

Before the dot: what kind of machine. After the dot: how big. That is the whole shape of every name.

m7iwhat kind
.
largehow big

The size part is the easy half. Everything doubles as you climb: large, xlarge, 2xlarge, 4xlarge. Twice the name, twice the machine.

One trap: large is the smallest, not a big one. It is the bottom of the ladder for almost every kind of machine.

02

Workers and desks

Picture a workshop. A server gives you two things: workers, who each do one job at a time, and desk space, which is what those workers can have spread out in front of them right now.

The workers are called vCPUs. Four workers means four jobs happen at once and a fifth waits. The desk is called memory, and it is very fast but it gets cleared completely every time the machine is switched off.

Now the point. The first letter of the name is just how much desk each worker gets. That is all it means.

c

Lots of workers, small desks

For jobs that are hard to think about but small to hold. AWS calls it compute optimized.

2 GiBper worker
m

An even split

The sensible default when you do not yet know which way your work leans. AWS calls it general purpose.

4 GiBper worker
r

Fewer workers, huge desks

For when the whole pile of data has to be spread out at once. AWS calls it memory optimized.

8 GiBper worker

A few other letters exist for machines built around something else entirely — big disks bolted inside (i, d), graphics chips for machine learning (g, p), or a network fast enough to run hundreds of machines as one (hpc). Same idea, different thing being favoured.

See these three side by side →
03 — optional

The letters in the middle

Everything between the first letter and the dot is detail. You can read most names without it. Come back when you need it.

The number, as in m7i

How new the hardware is. Higher is newer, and newer is normally both faster and cheaper than what it replaced. If you are choosing and unsure, take the biggest number available.

The letter after it, as in m7i

Whose chip is inside. i is Intel, a is AMD, g is Graviton, which is AWS's own design. Old names have no letter here at all, because everything was Intel back then.

Graviton is a different kind of chip. Software built for Intel or AMD will not start on it until it is rebuilt.

Anything after that, as in m7id

Extra hardware bolted on. d adds a fast disk inside the machine, n adds network speed. A handful of rarer letters exist and the decoder below knows them all.

Everything on that d disk is erased when you stop the machine. Use it for scratch, never as the only copy.

04

Try any name

This works from the rules, not from a list, so it can read names this page has never mentioned.

05 — reference

Every letter, in one place

The sections above cover the handful you will actually meet. This is the complete set, taken from AWS's own naming reference, for when you hit one you have not seen before. Not required reading.

One difference worth knowing: AWS lists all of the middle letters together in a single column called Options. This page splits them into two — whose silicon, and what extra hardware — because they answer different questions. The letters are identical either way.

06

Names that break the rules

A few genuinely do not follow the pattern. Worth knowing so you do not think the decoder is broken.

m5.largeNo chip letter at all. Older machines were all Intel, so the letter was never there to begin with.
g5g.xlargeThe letter g shows up twice meaning two different things — graphics machine, then Graviton chip. Only its place in the name tells them apart.
u7in-32tb.224xlargeThe amount of memory is written into the name itself. Unique to the enormous U machines.
mac2.metalA real Apple computer in a rack, rented whole. There is only one size because you get the entire thing.

That is the whole scheme. Next: what those ratios actually look like as hardware.

Open the explorer →

Explorer

What is actually inside one of these.

Every EC2 machine is built from the same set of parts. Once you can name them, the letters in the name stop being arbitrary — each one is turning a part up, down, or on.

The parts of an EC2 instance A server chassis containing workers, memory, a graphics chip, a local disk, a Nitro card and a network card, connected over the network to an EBS volume and to other machines. ONE INSTANCE — WHAT YOU RENT ELSEWHERE IN THE DATACENTRE 1 Workers · vCPU 2 Desk · memory 3 GPU 4 Disk inside 5 Nitro card 6 Network 7 EBS volume 8 Other machines

Every letter in an instance name turns one of these eight parts up, down, or on. That is the whole game.

Compare

Three machines, one difference at a time

Each comparison changes exactly one thing and holds everything else still, so you can see what that part of the name does to the hardware.

drag to look around

One block is one worker One module is 4 GiB of desk Disk inside the machine Network card Graphics chip
Size

Reference

One of every kind

The comparisons above teach the pattern. This is the coverage: one representative machine from each family, so you can see what actually exists and who makes the chips inside.

Built for
Chip

Specs are shown for one representative size and are correct as of writing — every family has a ladder of other sizes. If you fork this, regenerate the numbers rather than trusting them: aws ec2 describe-instance-types returns vCPU, memory, network and accelerator data for every type in a region.

Not sure what a worker or a desk is in real terms?

Back to the overview →