
Introduction
Plenty of production teams still run gp2 volumes that gp3 could serve for less. The hesitation is understandable: storage changes sound like downtime and risk, and nobody wants to explain an outage over a few hundred dollars a month.
This guide covers current US-region gp3 pricing, a cost model for comparing gp2 and gp3 fairly, and a step-by-step migration with AWS Elastic Volumes. Prices below are as of November 18, 2026, and they vary by Region and change over time.
"Without downtime" also depends on your volume, instance, operating system, and application, so check the prerequisites first.
Key Takeaways
- gp3 separates capacity from IOPS and throughput, so you size performance independently.
- Compare total cost: storage, extra IOPS, extra throughput, snapshots, and related charges.
- Supported volumes can be modified in place, but monitor until the change finishes.
- Baseline first, pilot one volume, then roll out in controlled batches.
What Are gp3 Volumes and How Is gp3 Pricing Calculated?
Amazon EBS gp3 is a General Purpose SSD volume for persistent EC2 block storage. AWS lists boot volumes, development and test environments, interactive applications, virtual desktops, and medium-sized single-instance databases as typical fits, according to the AWS gp3 documentation.
The three billing dimensions
- Provisioned storage, billed per GB-month
- Included baseline of 3,000 IOPS and 125 MiB/s
- Extra IOPS and throughput you provision above that baseline
All three are billed per second with a 60-second minimum.
Current US rates
| Region | gp3 $/GB-month | Extra IOPS $/IOPS-month | Extra throughput $/unit-month | gp2 $/GB-month |
|---|---|---|---|---|
| us-east-1 (N. Virginia) | 0.080 | 0.005 | 0.040 | 0.100 |
| us-east-2 (Ohio) | 0.080 | 0.005 | 0.040 | 0.100 |
| us-west-1 (N. California) | 0.096 | 0.006 | 0.048 | 0.120 |
| us-west-2 (Oregon) | 0.080 | 0.005 | 0.040 | 0.100 |
These come from the AWS EBS pricing page. AWS labels throughput in MB/s on that page and MiB/s in technical documentation; confirm the unit on the pricing page before you budget.
A reusable cost formula
Monthly cost = (GB × storage rate) + (billable IOPS × IOPS rate) + (billable throughput × throughput rate)
Billable IOPS is IOPS above 3,000. Billable throughput is anything above 125 MiB/s. For a partial month, multiply by the fraction of the month the volume existed.
Worked example in us-east-1: a 500 GB volume provisioned at 4,000 IOPS and 125 MiB/s.
- Storage: 500 × $0.080 = $40.00
- Extra IOPS: 1,000 × $0.005 = $5.00
- Extra throughput: none
- Total: $45.00/month
Costs that sit outside the volume charge
- Standard snapshots: $0.050/GB-month (us-west-1: $0.055)
- Snapshot Archive: $0.0125/GB-month, with a 90-day minimum
- Fast Snapshot Restore: $0.75 per DSU-hour per snapshot per AZ (us-west-1: $0.90)
- Cross-Region copies: data transfer plus destination snapshot storage
- Unattached volumes: billed from creation until deleted
- EBS direct APIs: billed per request, separate from normal I/O

gp2 vs. gp3: Cost and Performance Comparison
gp2 ties baseline performance to size: 3 IOPS per GiB, with a 100 IOPS minimum and 16,000 maximum. Volumes under 1 TiB can burst to 3,000 IOPS using credits.
gp3 has no credit model. You get 3,000 IOPS and 125 MiB/s from the start and pay only for what you add.
| Dimension | gp2 | gp3 |
|---|---|---|
| Capacity | 1 GiB to 16 TiB | 1 GiB to 64 TiB |
| Baseline IOPS | 3 per GiB (100 to 16,000) | 3,000 included, up to 80,000 |
| Bursting | Credits, volumes under 1 TiB | None needed |
| Throughput | 128 to 250 MiB/s | 125 included, up to 2,000 MiB/s |
| Pricing | Capacity only, I/O included | Capacity plus extra IOPS and throughput |
Maximum performance has prerequisites. 80,000 IOPS needs at least 160 GiB, and 2,000 MiB/s needs at least 8,000 IOPS and 16 GiB.
How to compare apples to apples
- Keep capacity identical.
- Set gp3 IOPS and throughput to match what gp2 actually delivers.
- Price each volume monthly.
- Add any performance charges needed for gp3 to match.
The 20% headline is a storage-rate comparison only. In AWS Prescriptive Guidance, a 500 GiB volume at 3,000 IOPS and 250 MiB/s models at $50 on gp2 versus $45 on gp3. That is 10% savings after throughput charges.
Smaller volumes that need more than their gp2 baseline benefit most. A 200 GiB gp2 volume has only 600 baseline IOPS, while gp3 gives 3,000 immediately.

Cases that need extra review
- Workloads that rely on gp2 burst behavior
- Volumes capped by EC2 instance EBS limits
- Applications needing more than gp3 maximums; AWS points these toward io2 Block Express
- Data better suited to st1, sc1, or another storage service
How to Migrate from gp2 to gp3 Without Downtime
Step 1: Inventory and baseline
Record each volume's ID, Region and AZ, attached instance, root or data role, size, current IOPS and throughput, filesystem, owner, and backup policy.
Then capture baselines from CloudWatch:
VolumeReadOpsandVolumeWriteOps(divide the sum by interval seconds for IOPS)VolumeReadBytesandVolumeWriteBytesVolumeQueueLengthBurstBalancefor gp2- Instance-level
EBSReadOps,EBSWriteOps, and byte metrics
Add application latency and database metrics. Size gp3 from measured peaks, not guesses.

Step 2: Snapshot and plan
Confirm a recent snapshot exists. Write down the owner, monitoring window, rollback triggers, and communication path. A snapshot captures persisted data, not unwritten cache, so it complements application-level backup and does not replace it.
Step 3: Modify the volume
In the console: EC2, Volumes, select the volume, Actions, Modify volume. Choose gp3, set IOPS and throughput, review the estimate, and submit.
With the CLI, per the AWS CLI reference:
aws ec2 modify-volume --volume-id VOLUME_ID --volume-type gp3 --iops IOPS --throughput MIBPS
--volume-type gp3sets the target type--iopsand--throughputset provisioned performance--size NEW_SIZE_GIBis optional and only increases capacity
The values are placeholders, not sizing advice.
What happens during modification
The volume stays attached and moves through modifying, optimizing, and completed. During optimization, performance sits between the old and new configurations. Large volumes can take hours, so keep monitoring.
Know the rules:
- A previous modification must finish first
- Four modifications are allowed per rolling 24 hours
- Requests can't be canceled
- Capacity can't be reduced in place
- Volumes attached before November 3, 2016 may need a detach/reattach or stop/start to enable modification
Changing type doesn't grow the filesystem. If you increase size, extend the partition and filesystem separately.

Validation, Rollback, and Operational Safeguards
Post-change checklist
- Confirm type, size, IOPS, and throughput with
aws ec2 describe-volumes-modifications --volume-ids VOLUME_ID - Check the attachment and modification state
- Check volume status is
ok - Compare CloudWatch metrics against baseline
- Review application health, database latency, and error rates
- Watch cost and usage data over the next billing cycle
Common causes of poor performance
- Unsupported Elastic Volumes configuration
- EC2 instance bandwidth ceilings
- IOPS or throughput set too low
- Filesystem or application limits
- Too few outstanding I/O requests
- Modification still optimizing
Rollback planning
Set measurable triggers before you start, such as p99 latency above an agreed threshold. Retain the pre-change snapshot according to your retention policy.
Two rollback paths exist:
- Revert the volume type with another
modify-volume --volume-type gp2request, subject to the same completion and rate limits - Restore data by creating a new volume from the snapshot and swapping it in; this requires detach and attach, so it is not an online change
Restored volumes can show first-access latency unless you initialize them or use Fast Snapshot Restore.
Ongoing EBS Cost Optimization After Migration
Migration isn't a one-time win. Revisit gp3 size, IOPS, and throughput against real CloudWatch usage. These AWS tools help you keep that review continuous:
- Cost Explorer shows spend trends, and activated cost allocation tags let you filter by team or app
- AWS Budgets alerts on actual or forecast overruns
- AWS Pricing Calculator estimates changes before you make them
- Compute Optimizer recommends type, size, IOPS, and throughput
- Trusted Advisor flags unattached or low-activity volumes
- Data Lifecycle Manager automates snapshot retention and deletion
Recurring waste usually hides in unattached volumes, storage on stopped instances, and obsolete snapshots. Ask whether cold data belongs on EBS at all.
Manual reviews don't scale across hundreds of volumes. Lucidity gives enterprise ITOps, DevOps, and FinOps teams real-time visibility into block storage across AWS, Azure, and Google Cloud.
Its Lumen product identifies unattached, reserved, unmounted, and zero-I/O disks that can represent up to 70% of unused block-storage spend. AutoScaler can then expand and shrink disks automatically. Use it alongside the migration validation above, not in place of it.

Frequently Asked Questions
What are gp3 volumes in Amazon EBS?
gp3 is a General Purpose SSD EBS volume type. Capacity, IOPS, and throughput are provisioned independently, unlike gp2, where performance follows size.
What is the price of Amazon EBS gp3?
In us-east-1, us-east-2, and us-west-2, storage is $0.080/GB-month, and us-west-1 is $0.096. It includes 3,000 IOPS and 125 MiB/s, with extra performance billed separately. Rates change, so check AWS pricing.
What is the difference between gp2 and gp3 AWS?
gp2 links baseline IOPS to capacity and uses burst credits. gp3 includes a flat baseline and lets you provision more. See the equivalent-performance comparison above for real costs.
Can you change an EBS volume from gp2 to gp3 without downtime?
Usually yes. Elastic Volumes modifies supported attached volumes in place with no detach or restart. Verify instance and volume support, monitor the modification, and validate performance.
How do you verify an EBS volume after migrating from gp2 to gp3?
Confirm type, IOPS, and throughput, and check that the modification state is completed. Compare CloudWatch metrics, application latency, and error rates to baseline, then review the next bill.