EC2 Instance Types: Which Family Should You Choose - M, C, R, or T?

Key Takeaway: EC2 Instance Type Selection

Bottom line: Start with M series (general purpose). Choose C series for CPU-intensive workloads, R series for memory-intensive workloads, and T series when you need burstable performance.

Quick Comparison

Family	Type	Characteristics	Primary Use Cases
M	General Purpose	Balanced CPU/memory	Web servers, business apps
T	Burstable	Baseline + burst	Dev/test, small websites
C	Compute Optimized	High CPU performance	Batch processing, HPC, gaming
R	Memory Optimized	Large memory	In-memory DB, caching
I/D	Storage Optimized	High-speed storage	NoSQL, data warehousing
P/G	Accelerated Computing	GPU/accelerators	ML/DL, graphics rendering

---

EC2 Instance Naming Convention

Understanding instance type names makes selection easier.

        m7i.2xlarge
        │││  │
        ││└──┴── Size (determines vCPU/memory)
        │└────── Generation (higher = newer, better value)
        └─────── Family (determines purpose)

Additional attribute examples:
m7i    = Intel processor
m7a    = AMD processor
m7g    = Graviton (ARM) processor
m7gd   = Graviton + NVMe SSD
m7in   = Intel + enhanced networking

Size Resource Multipliers

Size	vCPU Multiple	Memory Multiple
nano	1	-
micro	1	-
small	1	1x
medium	1	2x
large	2	1x
xlarge	4	2x
2xlarge	8	4x
4xlarge	16	8x

---

General Purpose Instances (M, T Series)

M Series: Fixed Performance General Purpose

The most common instances providing balanced compute, memory, and networking.

M Series Characteristics
┌─────────────────────────────────────────────────────┐
│         Balanced Resource Distribution              │
│                                                     │
│  ┌─────────┐  ┌─────────┐  ┌─────────┐             │
│  │  vCPU   │  │ Memory  │  │ Network │             │
│  │  ████   │  │  ████   │  │  ████   │             │
│  │  ████   │  │  ████   │  │  ████   │             │
│  └─────────┘  └─────────┘  └─────────┘             │
│                                                     │
│  • Good starting point when workload is unknown    │
│  • Baseline before switching to other families     │
└─────────────────────────────────────────────────────┘

Series	Processor	Features
M7i	Intel Sapphire Rapids	Latest Intel, high performance
M7a	AMD EPYC	Cost-effective
M7g	AWS Graviton3	Best price-performance

M Series Best For:

• Web application servers
• Business applications
• Small to medium databases
• Development and test environments
• Backend servers

T Series: Burstable General Purpose

Cost-effective instances offering baseline performance + burst capability.

T Series Burst Model
┌─────────────────────────────────────────────────────┐
│                                                     │
│  CPU Usage                                          │
│    ▲                                               │
│  100%│        ████                                 │
│     │  ████  █  ██                                 │
│  Base│──────────────────────────────────           │
│ (e.g.20%)                                          │
│     └──────────────────────────────────▶ Time      │
│                                                     │
│  CPU Credits                                        │
│    ▲                                               │
│     │    ████████                                  │
│     │  ██        ██████                            │
│     └──────────────────────────────────▶ Time      │
│                                                     │
│  • Accumulate credits during low usage             │
│  • Consume credits when bursting                   │
│  • Limited to baseline when credits depleted       │
└─────────────────────────────────────────────────────┘

Series	Processor	Features
T3	Intel	Most common
T3a	AMD	~10% cheaper than T3
T4g	Graviton2	40% better price-performance than T3

T Series Best For:

• Development and test environments
• Small websites
• Microservices
• Code repositories
• Low-traffic applications

---

Compute Optimized Instances (C Series)

Instances with high-performance processors for CPU-intensive workloads.

C Series Characteristics
┌─────────────────────────────────────────────────────┐
│         CPU-Centric Resource Distribution           │
│                                                     │
│  ┌─────────┐  ┌─────────┐  ┌─────────┐             │
│  │  vCPU   │  │ Memory  │  │ Network │             │
│  │  ████   │  │  ██     │  │  ████   │             │
│  │  ████   │  │  ██     │  │  ████   │             │
│  │  ████   │  │         │  │         │             │
│  │  ████   │  │         │  │         │             │
│  └─────────┘  └─────────┘  └─────────┘             │
│                                                     │
│  • Lower memory-to-vCPU ratio than M series        │
│  • Highest level CPU performance                   │
└─────────────────────────────────────────────────────┘

Series	Processor	Features
C7i	Intel Sapphire Rapids	Latest Intel
C7a	AMD EPYC	AVX-512 support
C7g	Graviton3	Best price-performance
C7gn	Graviton3 + enhanced networking	Up to 200Gbps

C Series Best For:

• Batch processing
• Media transcoding
• High-performance web servers
• High-performance computing (HPC)
• Scientific modeling
• Dedicated game servers
• Ad serving engines
• Machine learning inference

---

Memory Optimized Instances (R, X, Z Series)

Instances with large memory for memory-intensive workloads.

R/X Series Characteristics
┌─────────────────────────────────────────────────────┐
│         Memory-Centric Resource Distribution        │
│                                                     │
│  ┌─────────┐  ┌─────────┐  ┌─────────┐             │
│  │  vCPU   │  │ Memory  │  │ Network │             │
│  │  ██     │  │  ████   │  │  ███    │             │
│  │  ██     │  │  ████   │  │  ███    │             │
│  │         │  │  ████   │  │         │             │
│  │         │  │  ████   │  │         │             │
│  └─────────┘  └─────────┘  └─────────┘             │
│                                                     │
│  • R: 16 GiB at large (M series is 8 GiB)         │
│  • X: Even larger memory (up to 4 TiB)            │
│  • Z: High-frequency CPU + large memory            │
└─────────────────────────────────────────────────────┘

Series	Features	Max Memory
R7i	Intel, general memory optimization	1,024 GiB
R7g	Graviton3, best value	512 GiB
X2idn	Ultra-large memory	2,048 GiB
X2iedn	Ultra-large + NVMe	4,096 GiB
z1d	High-frequency CPU + memory	384 GiB

R/X/Z Series Best For:

• In-memory databases (SAP HANA, Redis)
• In-memory caching (ElastiCache)
• Real-time big data analytics
• Large enterprise applications
• High-performance relational databases

---

Storage Optimized Instances (I, D Series)

Instances with high-speed local storage for storage-intensive workloads.

Series	Features	Use Cases
I4i	NVMe SSD, high IOPS	NoSQL DB, OLTP
I4g	Graviton2 + NVMe	Cost-effective NoSQL
D3	HDD, high-density storage	Data warehousing
D3en	Ultra-large HDD	Big data analytics

I/D Series Best For:

• NoSQL databases (Cassandra, MongoDB)
• Data warehousing
• Distributed file systems (HDFS)
• Elasticsearch, log analytics

---

Accelerated Computing Instances (P, G, Inf, Trn Series)

Instances with GPUs and AI accelerators for specialized workloads.

Series	Accelerator	Use Cases
P5	NVIDIA H100	Large-scale ML/DL training
P4d	NVIDIA A100	ML/DL training
G5	NVIDIA A10G	Graphics rendering, ML inference
G6	NVIDIA L4	Cost-effective graphics
Inf2	AWS Inferentia2	ML inference specialized
Trn1	AWS Trainium	DL training specialized

Accelerated Computing Best For:

• Deep learning model training (P series)
• Machine learning inference (Inf series)
• 3D graphics rendering (G series)
• Video transcoding
• Game streaming

---

Processor Comparison: Intel vs AMD vs Graviton

Processor Selection Guide

Processor	Advantages	Disadvantages	Recommended For
Intel (i)	Wide compatibility, stability	Relatively expensive	Legacy app migration
AMD (a)	~10% cheaper than Intel	Check some app compatibility	Cost optimization
Graviton (g)	Up to 40% better price-performance	ARM architecture compatibility	New workloads, containers

Graviton Considerations

Graviton Compatible:
✅ Container-based applications
✅ Open-source software
✅ Java, Python, Node.js applications
✅ Web servers (NGINX, Apache)
✅ Databases (MySQL, PostgreSQL)

Graviton - Verify First:
⚠️ Windows-only applications
⚠️ x86-only binaries
⚠️ Special hardware drivers

---

Instance Type Selection Guide

Decision Flowchart

What's your workload characteristic?
    │
    ├── CPU-intensive ──▶ C series
    │
    ├── Memory-intensive ──▶ R/X series
    │
    ├── Storage-intensive ──▶ I/D series
    │
    ├── GPU/ML required ──▶ P/G/Inf/Trn series
    │
    └── Balanced needed ──▶ What's the CPU usage pattern?
                              │
                              ├── Consistently high ──▶ M series
                              │
                              └── Intermittent ──▶ T series

Recommended Instances by Workload

Workload	Recommended Family	Recommended Series
Web server	General Purpose	M7i, M7g, T3
API server	General Purpose	M7i, M7g
Batch processing	Compute	C7i, C7g
Media encoding	Compute	C7i, C6i
In-memory cache	Memory	R7i, R7g
SAP HANA	Memory	X2idn, R7iz
NoSQL DB	Storage	I4i, I4g
ML training	Accelerated	P5, P4d, Trn1
ML inference	Accelerated	Inf2, G5
Dev/test	Burstable	T3, T4g

---

Cost Optimization

Instance Cost Reduction Strategies

1. Select the Right Family

   • Avoid over-provisioning for your workload
   • Start with general purpose → switch as needed

2. Leverage Graviton

   • Save up to 40% with compatible workloads

3. Use Latest Generation

   • Better performance at the same cost
   • Example: M6i → M7i (15% performance improvement)

4. Combine Purchase Options

   • Stable workloads: Reserved Instances
   • Flexible workloads: Savings Plans
   • Interruptible workloads: Spot Instances

---

SAA-C03 Exam Focus Points

Common Question Types

1. Instance Selection by Workload

   • "Batch processing, CPU-intensive" → C series
   • "In-memory database" → R series
   • "Large-scale ML training" → P series

2. T Series Burst Credits

   • "CPU credits depleted" → Switch to M series or enable Unlimited mode

3. Cost Optimization

   • "Price-performance, cost savings" → Graviton (g suffix)
   • "Predictable workloads" → Reserved/Savings Plans

4. SAP HANA Instances

   • Large memory required → X series or R series

Key Memorization Points

Keyword	Association
Balanced, general purpose	M series
Burstable, variable CPU	T series
CPU-intensive, batch	C series
Memory-intensive, in-memory DB	R series
Ultra-large memory, SAP	X series
High-speed storage, NoSQL	I series
ML training, GPU	P series
ML inference	Inf series
Price-performance, ARM	Graviton (g)

---

Frequently Asked Questions

Q: Should I choose M series or T series?

A: Choose based on your CPU usage pattern:

• M series: When CPU usage is consistently high (e.g., 24/7 production servers)
• T series: When CPU usage is intermittent (e.g., dev servers, small websites)

Q: Can I use Graviton instances?

A: Most Linux-based open-source workloads are compatible with Graviton. Container, Java, and Python applications typically run without major modifications. However, verify x86-only binaries first.

Q: Can I change instance types?

A: Yes, you can change EC2 instance types after stopping the instance. Generally, changes within the same hypervisor (Nitro/Xen) are straightforward.

Q: Which generation should I choose?

A: Choose the latest generation (7th gen or newer) when possible. Newer generations provide better performance and features at the same cost.

Q: Storage Optimized (I/D) vs EBS io2 - which should I choose?

A:

• I/D series: When ultra-fast local storage is needed and data persistence is less critical
• EBS io2: When data persistence is required, snapshots/replication needed

---

Summary

The key to EC2 instance type selection is understanding your workload characteristics:

1. Starting Point: Begin with M series (general purpose)
2. Optimization: Monitor and switch to appropriate family
3. Cost Savings: Leverage Graviton, choose latest generation

On the SAA-C03 exam, matching workload keywords to instance families is frequently tested. Remember: "Batch processing" → C, "In-memory" → R, "ML training" → P.