Skip to content

Latest commit

 

History

History
215 lines (104 loc) · 4.93 KB

aws-instance-types.md

File metadata and controls

215 lines (104 loc) · 4.93 KB

AWS Instance Type Cheat Sheet

Cache invalidation and naming things - AWS has certainly had a hard time of the latter. Between the variety of instance types they offer and how long they're been around, there’s no consistency.

Whether starting a new service to run internally or reviewing something we have going, it’s always a struggle to find the right instance type for the needs. For example, there are three families (r, x, z) that optimize RAM in various ways in various combinations yet I always forget about the x and z variants.

So I put together this "cheat sheet" for us internally and thought I'd share here for anyone else to use; pull requests welcome for updates.

NOTE: Each vCPU is a thread of an Intel or AMD core, except where noted.

Family Overview

  • a: First generation graviton Arm processors, use m6g, c6g, or r6g instead.
  • c: Compute-optimized with various core types.
  • d: Locally attached spinning HDD.
  • f: FPGAs to run ASICs.
  • g: GPUs for graphics or machine-learning.
  • h: Large spinning HDD.
  • i: High-throughput local SSD.
  • m: General-purpose with balanced memory, CPU, and storage.
  • r: Memory-optimized.
  • t: Burstable instances that will throttle after a CPU budget.
  • x: Lowest price-per-GB RAM instances.
  • z: Highest core frequency with a lot of RAM.

A

First generation graviton Arm processors, use m6g, c6g, or r6g instead.

Variants

a1 64-bit Arm Neoverse cores

C

Compute-optimized with various core types.

Variants

c4 Intel Xeon E5-2666 v3 cores

c5 Xeon Platinum 8000 cores

c5a AMD EPYC 7002 cores

c5ad c5a + locally attached SSD

c5d c5 + locally attached SSD

c5n c5 + optimized network

c6g Arm-based AWS Graviton2 *vCPU is physical core

c6gd c6g with locally attached SSD storage

D

Locally attached spinning HDD.

Variants

d2 Attached spinning HDD

d3en High-throughput w/attached spinning HDD

F

FPGAs to run ASICs.

Variants

f1 Attached FPGA and SSD

G

GPUs for graphics or machine-learning.

Variants

g3 Attached NVIDIA Tesla M60 GPU

g4ad Attached AMD Radeon Pro V520 GPU

g4dn Attached NVIDIA T4 Tensor Core GPU

H

Large spinning HDD.

Variants

h1 Up to 16TB of spinning HDD.

I

High-throughput local SSD.

Variants

i3 Attached NVMe SSD

i3en Larger SSDs and enhanced networking

inf1 AWS Inferentia for machine learning

M

General-purpose with balanced memory, CPU, and storage.

Variants

m4 Balanced, older Intel Xeon cores, use m5 or m6g instead

m5 Balanced CPU and memory, latest Intel Xeon cores

m5a Balanced CPU and memory, latest AMB EPYC cores

m5ad m5a with attached SSD

m5d m5 with attached SSD

m5n m5 with faster networking

m5zn m5n with highest 4.5GHz core speed

m6g Balanced CPU and memory, latest Amazon Graviton 2 cores

m6gd m6g with attached SSD storage

P

FAMILYDESC

Variants

p2 NVIDIA K80 GPUs

p3 NVIDIA Tesla V100 GPUs

p4 NVIDIA A100 Tensor GPUs

R

Memory-optimized.

Variants

r4 Older high memory, use r5 or r6g instead.

r5 Latest high memory, Intel Xeon cores

r5a Latest high memory, AMD EPYC cores

r5ad r5a with attached SSD

r5b r5 with optimized EBS

r5dn r5n with added SSD

r5n r5 with faster networking

r6g Latest high memory, Amazon Graviton 2 cores

r6gd r6g with attached SSD

T

Burstable instances that will throttle after a CPU budget.

Variants

t2 Older burstable instances *vCPU is physical core, use t3 or t4g instead

t3 Latest generation, burstable instances with Intel Xeon cores

t3a Latest generation, burstable instances with AMD EPYC cores

t4g Latest generation, burstable instances with Amazon Graviton 2 cores

X

Lowest price-per-GB RAM instances.

Variants

x1 Memory-optimized Intel Xeon E7-8880 v3

x1e x1 with optimized EBS

Z

Highest core frequency with a lot of RAM.

Variants

z1d Fast single core performance with really large CPU and memory configurations