gp2 to gp3 migration: cost savings with no performance loss
On the Amazon EBS pricing page (checked November 2026), gp3 costs less per GB-month than gp2 and includes 3,000 IOPS and 125 MiB/s on every volume. That makes most gp2 volumes cheaper and no slower after one live aws ec2 modify-volume --volume-type gp3 call. The only real work is setting IOPS and throughput on larger gp2 volumes, whose size-based limits of 3 IOPS per GiB and up to 250 MiB/s exceed what gp3 includes.
Why is gp2 more expensive than gp3?
gp2 ties performance to size. A gp2 volume gets 3 IOPS per GiB, with a minimum of 100 and a cap of 16,000 at 5,334 GiB. Volumes under 1,000 GiB can burst to 3,000 IOPS on credits. Throughput reaches 250 MiB/s at 334 GiB and above.
gp3 prices storage lower per GB and includes 3,000 IOPS and 125 MiB/s on every volume, whatever its size. You pay extra only for IOPS and throughput you provision above that baseline. The ceilings are 80,000 IOPS and 2,000 MiB/s (raised from 16,000 and 1,000 in September 2025), within two caps: 500 IOPS per GiB of volume size, and 0.25 MiB/s per provisioned IOPS.
AWS EBS pricing (https://aws.amazon.com/ebs/pricing/, checked November 2026): gp3 storage costs less per GB-month than gp2. gp3 adds two line items only when you provision above the baseline: a per-IOPS-month charge above 3,000 IOPS and a per-MiB/s-month charge above 125 MiB/s. Rates differ by region, so read the rows for the regions on your bill, or take your effective rate from your Cost and Usage Report.
# List every gp2 volume in the current region with size and IOPS
aws ec2 describe-volumes --filters Name=volume-type,Values=gp2 --query 'Volumes[].[VolumeId,Size,Iops,State]' --output table
How do I find gp2 vs. gp3 costs in Cost Explorer?
In Cost Explorer (Billing and Cost Management > Cost Explorer), EBS volume charges sit under Service = "EC2-Other", not a separate EBS service. Group by Usage type. Storage shows as EBS:VolumeUsage.gp2 and EBS:VolumeUsage.gp3, and gp3 performance above the baseline shows as EBS:VolumeP-IOPS.gp3 and EBS:VolumeP-Throughput.gp3.
Outside us-east-1, usage types carry a region prefix, such as USW2-EBS:VolumeUsage.gp2. Cost Explorer stops at usage type. For cost per volume, query the Cost and Usage Report in Athena with resource IDs enabled.
-- Athena on the CUR: last month's gp2/gp3 cost by usage type and volume
-- Needs resource IDs in the CUR; replace cur_db.cur_table and the dates
SELECT
line_item_usage_type,
line_item_resource_id,
SUM(line_item_usage_amount) AS usage_amount,
SUM(line_item_unblended_cost) AS cost
FROM cur_db.cur_table
WHERE line_item_product_code = 'AmazonEC2'
AND line_item_line_item_type = 'Usage'
AND line_item_usage_type LIKE '%EBS:Volume%'
AND (line_item_usage_type LIKE '%.gp2' OR line_item_usage_type LIKE '%.gp3')
AND line_item_usage_start_date >= TIMESTAMP '2026-10-01 00:00:00'
AND line_item_usage_start_date < TIMESTAMP '2026-11-01 00:00:00'
GROUP BY 1, 2
ORDER BY cost DESC;
Check your own bill in 2 minutes: https://revelations.consulting/checker?utm_source=blog&utm_medium=organic&utm_content=blog-gp2-to-gp3-migration
How do I migrate from gp2 to gp3 without downtime?
Run aws ec2 modify-volume with --volume-type gp3, and pass --iops and --throughput in the same call, using the values you take from CloudWatch in the next section. If you leave them out, AWS sets gp3's defaults of 3,000 IOPS and 125 MiB/s. The volume stays attached and in use throughout. Track progress with aws ec2 describe-volumes-modifications; the state moves from modifying to optimizing to completed.
# Convert and set performance in one call: one modification, one six-hour window
aws ec2 modify-volume --volume-id vol-xxxxxxxxxxxxx --volume-type gp3 --iops 3000 --throughput 125
# Track progress: modifying -> optimizing -> completed
aws ec2 describe-volumes-modifications --volume-ids vol-xxxxxxxxxxxxx --query 'VolumesModifications[0].[VolumeId,ModificationState,Progress]' --output table
No instance reboot is required. While the state is optimizing, the volume delivers performance somewhere between its gp2 and gp3 specs. Optimizing can take from minutes to several hours on a large volume, so convert busy volumes in a quiet period.
What IOPS and throughput settings should I use?
Before you convert, pull 14 days of per-minute CloudWatch data for the gp2 volume. VolumeReadOps plus VolumeWriteOps, divided by the period in seconds, gives IOPS. VolumeReadBytes plus VolumeWriteBytes, divided the same way, gives throughput. Size gp3 from the peaks, not the averages.
# Peak IOPS and MiB/s for one volume over 14 days (GNU date)
aws cloudwatch get-metric-data \
--start-time "$(date -u -d '14 days ago' +%Y-%m-%dT%H:%M:%SZ)" \
--end-time "$(date -u +%Y-%m-%dT%H:%M:%SZ)" \
--metric-data-queries '[
{"Id":"r","MetricStat":{"Metric":{"Namespace":"AWS/EBS","MetricName":"VolumeReadOps","Dimensions":[{"Name":"VolumeId","Value":"vol-xxxxxxxxxxxxx"}]},"Period":60,"Stat":"Sum"},"ReturnData":false},
{"Id":"w","MetricStat":{"Metric":{"Namespace":"AWS/EBS","MetricName":"VolumeWriteOps","Dimensions":[{"Name":"VolumeId","Value":"vol-xxxxxxxxxxxxx"}]},"Period":60,"Stat":"Sum"},"ReturnData":false},
{"Id":"rb","MetricStat":{"Metric":{"Namespace":"AWS/EBS","MetricName":"VolumeReadBytes","Dimensions":[{"Name":"VolumeId","Value":"vol-xxxxxxxxxxxxx"}]},"Period":60,"Stat":"Sum"},"ReturnData":false},
{"Id":"wb","MetricStat":{"Metric":{"Namespace":"AWS/EBS","MetricName":"VolumeWriteBytes","Dimensions":[{"Name":"VolumeId","Value":"vol-xxxxxxxxxxxxx"}]},"Period":60,"Stat":"Sum"},"ReturnData":false},
{"Id":"iops","Expression":"(r+w)/PERIOD(r)","Label":"IOPS"},
{"Id":"mibps","Expression":"(rb+wb)/PERIOD(rb)/1048576","Label":"MiB/s"}
]' \
--query 'MetricDataResults[].[Label,max(Values)]' --output table
If both peaks stay under 3,000 IOPS and 125 MiB/s, gp3's included baseline covers the volume. If the gp2 volume's own limits exceed gp3's baseline, set gp3 to at least the observed peak plus headroom. A 5,334 GiB gp2 volume, for example, has 16,000 IOPS and 250 MiB/s. If your data is thin, match the gp2 figures exactly.
# Large gp2 volume (5,334 GiB = 16,000 IOPS, 250 MiB/s): match it on gp3 in the same call as the type change
aws ec2 modify-volume --volume-id vol-xxxxxxxxxxxxx --volume-type gp3 --iops 16000 --throughput 250
Do not assume gp3's defaults are enough. A gp2 volume that ran at 12,000 IOPS or 250 MiB/s will slow down on gp3 left at 3,000 IOPS and 125 MiB/s.
What can go wrong when I switch to gp3?
AWS requires at least six hours between modifications on the same volume. If you set IOPS or throughput too low, you live with it until the window passes. So set both in the conversion call, never as a follow-up.
Start with a non-critical volume and take a snapshot first. Watch VolumeQueueLength and average latency (VolumeTotalWriteTime divided by VolumeWriteOps) for a few days, then convert the rest in batches.
gp3 is also not automatically cheaper. You pay for the IOPS and throughput you provision, not what you use, so price high settings on the EBS pricing page before you apply them.
What it usually looks like
Sort your gp2 inventory by size and check it against three bands.
Volumes under 1,000 GiB, such as boot and small app volumes, burst to at most 3,000 IOPS on gp2. gp3's included 3,000 IOPS covers them, so confirm throughput peaks stay under 125 MiB/s and convert at defaults. Volumes of 1,000 GiB and up have a gp2 baseline above 3,000 IOPS (3 per GiB, capped at 16,000 from 5,334 GiB), so set gp3 IOPS from each volume's observed peak.
The third band is throughput. Any gp2 volume of 334 GiB or more can deliver 250 MiB/s, and databases, Kafka brokers and backup targets use it during compactions, scans and snapshots. If VolumeReadBytes plus VolumeWriteBytes peaks above 125 MiB/s, set gp3 throughput to cover the peak. Otherwise the large sequential reads and writes will show latency spikes that look like an IOPS problem.
I cut a roughly $1M-a-year cloud bill by 26% at a former employer. gp2 conversions are where I would start on any bill, because they run live and need no code change.
Every gp2 volume costs more per GB than the same volume on gp3, so the saving applies to all of them. This week, pick one non-critical volume. Pull its 14-day peaks, convert it with explicit IOPS and throughput in a single call, and watch latency for a few days before you batch the rest.
You can do all of this yourself with the commands above. If you want senior engineers to look at the rest of your cloud and AI bill with you, book a free 30-minute consult: https://revelations.consulting/?utm_source=blog&utm_medium=organic&utm_content=blog-gp2-to-gp3-migration#contact
Frequently asked questions
How do I know if my gp2 volume is over-provisioned?
A gp2 volume is over-provisioned when it was sized up for IOPS it never uses. Compare its 14-day peak IOPS, (VolumeReadOps + VolumeWriteOps) divided by the period in seconds, with the gp2 baseline of 3 IOPS per GiB. If the peak is far below that baseline, set the gp3 replacement's IOPS to the peak plus headroom; EBS cannot shrink a volume in place, so reclaiming unused GiB means copying to a new, smaller volume.
Do I need to reboot the instance after migrating to gp3?
No. Converting an EBS volume from gp2 to gp3 with aws ec2 modify-volume happens while the volume stays attached and in use, with no reboot or detach. During the optimizing phase, which can last from minutes to several hours on large volumes, the volume's performance sits between its gp2 and gp3 specs.
Can I revert from gp3 back to gp2?
Yes. Run aws ec2 modify-volume --volume-type gp2 once at least six hours have passed since the volume's last modification and that modification has completed. There is rarely a reason to, since gp2 costs more per GB and ties IOPS to volume size.
What’s the maximum IOPS and throughput I can set on a gp3 volume?
A gp3 volume supports up to 80,000 IOPS and 2,000 MiB/s, limits AWS raised from 16,000 IOPS and 1,000 MiB/s in September 2025. gp3 IOPS are capped at 500 per GiB of volume size, and gp3 throughput at 0.25 MiB/s per provisioned IOPS, so small volumes cannot reach the maximums. Check the Amazon EBS volume types documentation for current limits before provisioning.
How do I know if I need to increase throughput on gp3?
Compute gp3 throughput from CloudWatch's VolumeReadBytes and VolumeWriteBytes: their sum divided by the period in seconds gives bytes per second. If peak throughput nears the gp3 volume's provisioned throughput, 125 MiB/s by default, during backups, compactions or large sequential reads and writes, raise it. A gp2 volume of 334 GiB or more could deliver 250 MiB/s, so match that figure when the gp2 peaks ran above 125 MiB/s.
Want this kind of thing weekly? Get Cloud Bill Weekly. Want us to find it in your bill? We're taking three founding clients for a free Cost Assessment.