EBS Volume Types: What They Are and When to Use Each

Definition

Amazon EBS offers six distinct volume types in 2026, each tuned for a specific balance of IOPS, throughput, durability, and cost. The SSD-backed types — gp3, gp2, io2 Block Express, and io1 — are optimized for low-latency random I/O and are the right choice for boot volumes and transactional databases. The HDD-backed types — st1 and sc1 — are optimized for large sequential throughput and low cost, ideal for big-data and infrequently accessed workloads. Picking the right volume type is one of the most common AWS architecture decisions, and it appears on nearly every AWS certification exam.

How It Works

All EBS volumes attach over the AWS network to a single EC2 instance in a specific Availability Zone (or to multiple Nitro instances via Multi-Attach for io1/io2 with cluster-aware filesystems). The EC2 hypervisor presents the volume as a NVMe or SCSI block device; the guest OS formats it and mounts a filesystem.

• SSD-backed volumes are measured in IOPS (input/output operations per second) and throughput (MB/s).
• HDD-backed volumes prioritize throughput (MB/s); IOPS are low because spinning disks excel at sequential access.
• Elastic Volumes let you change size, type, and IOPS without detaching the volume or stopping the instance (one modification per volume every 6 hours).
• gp3 uses fixed baseline performance with independent IOPS/throughput dials. gp2 uses a credit-based burst model tied to volume size.
• io1/io2 let you provision IOPS explicitly and pay separately.
• io2 Block Express is a next-generation architecture delivering sub-millisecond latency, up to 256,000 IOPS, 4,000 MB/s throughput, and 99.999% durability.

Key Features and Limits

Complete volume type comparison

| Type | Backing | Baseline | Max IOPS | Max Throughput | Durability (AFR) | Max Size | Typical Cost | | --- | --- | --- | --- | --- | --- | --- | --- | | gp3 | SSD | 3,000 IOPS + 125 MB/s | 16,000 | 1,000 MB/s | 99.8–99.9% | 16 TiB | $ | | gp2 (legacy) | SSD | 3 IOPS/GB (burst 3,000) | 16,000 | 250 MB/s | 99.8–99.9% | 16 TiB | $ | | io2 Block Express | SSD | Provisioned | 256,000 | 4,000 MB/s | 99.999% | 64 TiB | $$$ | | io1 (legacy) | SSD | Provisioned | 64,000 | 1,000 MB/s | 99.8–99.9% | 16 TiB | $$$ | | st1 | HDD | 40 MB/s per TiB | 500 | 500 MB/s | 99.8–99.9% | 16 TiB | ¢ | | sc1 | HDD | 12 MB/s per TiB | 250 | 250 MB/s | 99.8–99.9% | 16 TiB | ¢ |

gp3 details

Every gp3 volume starts with 3,000 IOPS baseline + 125 MB/s throughput, regardless of size. You can scale IOPS up to 16,000 and throughput up to 1,000 MB/s independently, each at an incremental per-unit charge. gp3 is AWS's recommended default — it delivers better baseline performance than gp2 and typically costs about 20% less.

io2 Block Express details

io2 Block Express runs on a new SRD-based storage fabric with sub-millisecond latency. It provides up to 256,000 IOPS, 4,000 MB/s throughput, a volume size up to 64 TiB, and 99.999% annual durability — the highest in EBS. Ideal for SAP HANA, Oracle RAC, and other mission-critical latency-sensitive databases.

HDD throughput

st1 sustains 40 MB/s per TiB baseline with 250 MB/s per TiB burst, capped at 500 MB/s. sc1 is 12 MB/s per TiB baseline, capped at 250 MB/s. HDD volumes cannot be boot volumes.

Multi-Attach

Only io1 and io2 support Multi-Attach (up to 16 Nitro instances in the same AZ), and only with cluster-aware filesystems like OCFS2 or GFS2.

Common Use Cases

1. Boot volumes — gp3 (AWS default; gp2 for legacy AMIs).
2. Transactional databases — gp3 for small-to-mid MySQL/PostgreSQL; io2 Block Express for SAP HANA, Oracle, SQL Server.
3. NoSQL on EC2 — Cassandra, Elasticsearch: gp3 for typical workloads; io2 for hot shards.
4. Big data & log ingestion — st1 for Hadoop/Spark DataNodes and Kafka log segments.
5. Cold archival stores — sc1 for infrequently accessed bulk data that must be EBS-backed.
6. Clustered apps — io2 Multi-Attach for Windows Failover Clustering or Oracle RAC.

Pricing Model

• gp3: ~$0.08/GiB-month (varies by Region). Above baseline, additional IOPS ~$0.005/IOPS-month, additional throughput ~$0.04/MB/s-month.
• gp2: ~$0.10/GiB-month, no provisioned IOPS charge (IOPS scale with size).
• io2: ~$0.125/GiB-month + tiered IOPS pricing (higher price per IOPS above 32k).
• io2 Block Express: same storage price tier, extended IOPS pricing; requires Nitro-based R5b/X2idn/etc. instance.
• io1: similar to io2, usually slightly cheaper but lower durability.
• st1: ~$0.045/GiB-month.
• sc1: ~$0.015/GiB-month.

All EBS volumes also incur per-GB-month snapshot storage when you back them up. Volumes continue billing when the attached EC2 instance is stopped, only ending when the volume is deleted.

Pros and Cons

Pros

• Broad spectrum of price/performance — from penny-per-GB HDD archives to 256k-IOPS SSDs.
• Elastic Volumes let you resize or change type online.
• gp3 decouples IOPS/throughput from capacity — pay only for what you need.
• io2 Block Express delivers mission-critical database performance without SAN hardware.

Cons

• All volumes are AZ-scoped — moving across AZs requires snapshot + restore.
• Provisioned capacity: you pay for empty space.
• HDD types have low IOPS (500 or 250 total) — avoid for random-access workloads.
• io2 Block Express is only supported by certain Nitro instance families (R5b, X2idn, X2iedn, M7i, etc.).
• Multi-Attach is restricted to io1/io2 with cluster filesystems — many production databases can't use it.

Comparison with Alternatives

| Need | EBS choice | Alternative | | --- | --- | --- | | Boot volume | gp3 | N/A | | 10k IOPS DB | gp3 | io2 if latency matters | | 100k+ IOPS DB | io2 Block Express | FSx for NetApp ONTAP | | Big-data throughput | st1 | S3 + Athena (for analytics) | | Shared filesystem | io2 Multi-Attach (cluster FS) | EFS, FSx | | Ephemeral scratch | Instance store | N/A | | Cold bulk | sc1 | S3 Glacier Instant Retrieval |

Exam Relevance

• Solutions Architect Associate (SAA-C03) — map workload descriptions to the right volume type. Memorize the gp3 baseline (3,000 IOPS + 125 MB/s), io2 Block Express max (256k IOPS, 4 GB/s, 99.999% durability), and HDD caps (st1: 500 IOPS / 500 MB/s; sc1: 250 / 250).
• Developer Associate (DVA-C02) — picking gp3 over gp2 for cost savings, understanding Elastic Volumes for zero-downtime resizing.
• SysOps Administrator (SOA-C02) — monitoring VolumeReadOps/VolumeWriteOps, burst-balance exhaustion (gp2), switching gp2 → gp3 for cost optimization, Data Lifecycle Manager scheduling.

Classic exam trap: the question describes a Hadoop cluster or Kafka ingestion — the answer is st1, not gp3. Another: "needs 99.999% durability" — this is io2 / io2 Block Express, not gp3 (which is 99.8–99.9%).

Frequently Asked Questions

Q: What is the difference between gp2 and gp3 EBS volumes?

A: Both are general-purpose SSDs with the same 16,000 IOPS ceiling, but gp3 provides a fixed 3,000 IOPS baseline + 125 MB/s throughput regardless of size, and lets you dial IOPS and throughput independently up to 16,000 / 1,000 MB/s. gp2 uses a credit-based burst model where IOPS scale with volume size (3 IOPS/GB). gp3 is typically ~20% cheaper for equivalent performance, making it AWS's current default and the recommended migration target for gp2 volumes.

Q: When should I use io2 Block Express over gp3?

A: Choose io2 Block Express when your workload needs more than 16,000 IOPS, sub-millisecond latency, throughput above 1 GB/s, or 99.999% durability — typical of SAP HANA, Oracle Database, large SQL Server OLTP, or tier-1 financial systems. For most workloads (web apps, CI/CD, smaller databases, boot volumes), gp3 is dramatically cheaper and fast enough.

Q: What EBS volume type is best for throughput-intensive workloads like Hadoop?

A: Use st1 — throughput-optimized HDD. It provides up to 500 MB/s per volume and is engineered for large sequential reads and writes, which matches Hadoop/Spark DataNodes, Kafka log segments, and batch log-processing pipelines. Its IOPS ceiling is only 500, so it is unsuitable for random-access workloads, but its price per GB is about half that of gp3 — a substantial saving at petabyte scale.

---

This article reflects AWS features and pricing as of 2026. AWS services evolve rapidly — always verify against the official EBS volume types documentation before making production decisions.