Migrate EBS volumes from gp2 to gp3

gp2 and gp3 are both general-purpose SSD volume types on AWS EBS — the storage that EC2 instances use for root disks and data volumes. gp2 is the original type; gp3 is the replacement AWS shipped in 2020 at roughly 20% lower cost per GB. That's not a promotional rate or a tiered discount — it's the standard price, available today, for every volume in every region.

The cost difference is clean: gp2 bills at ~$0.10/GB-month and gp3 at ~$0.08/GB-month. On a 10 TB EBS footprint that's $205/month — $2,460/year — recurring forever with no further action. The savings compound further because gp2 forces teams to provision oversized volumes just to get enough IOPS (every GB buys 3 IOPS), so the fleet is often larger than it needs to be. gp3 decouples IOPS from size and bundles 3,000 baseline IOPS at no extra cost, which means a volume that was 1 TB just for IOPS headroom can be right-sized to 200 GB on gp3.

There is no lock-in risk here: no Reserved Instance commitment, no snapshot prerequisite, no maintenance window. The migration is a single API call against a live volume. FinOps scanners surface gp2 as a finding precisely because it is low-risk, immediately actionable, and leaves real money on the table every month it isn't done.

Raj is the FinOps lead for a SaaS company. The monthly cloud cost review surfaces a $14,200/month EBS line, and Cost Optimization Hub flags the entire line as gp2 — 142 TB of volumes across two regions, untouched since the original cloud migration three years ago.

Raj doesn't need to understand the API to own this number. He pulls the per-GB math: 142 TB at ~$0.10 is $14,200; at ~$0.08 it's $11,360 — a $2,840/month delta, $34,080/year. He also asks engineering whether any of those volumes are oversized for IOPS reasons; the answer is yes, roughly 30% of the volume capacity was provisioned purely for gp2 IOPS scaling. That suggests a combined saving closer to $40,000/year once right-sizing follows the type migration.

Raj adds a migration milestone to the quarterly FinOps roadmap with a target: zero gp2 volume-hours on the bill by end of quarter. He tracks progress each week in Cost Explorer — the gp2 GB-month line should fall to zero while the gp3 line rises by the same total capacity. The engineer does the API work; Raj owns the outcome metric and reports the realized saving against the forecast in the next finance review.

The visible cost is the 20% per-GB premium. At $0.10/GB-month versus $0.08/GB-month, a 10 TB gp2 footprint costs $2,460/year more than the same capacity on gp3. That number compounds silently — it has been accruing since the original deployment, every month, with no notification. Most finance teams don't see it because EBS storage rarely gets a line of its own; it hides inside the EC2 or catch-all storage charge.

The second cost is hidden overcapacity. gp2 forces teams to provision extra gigabytes they don't need for storage just to get IOPS — 3 IOPS per GB means you need a 1,000 GB volume to get 3,000 IOPS, even if the workload only uses 200 GB. On gp3 those 3,000 IOPS come free with any volume size, so the correct volume is 200 GB. A finance partner who only models the per-GB rate cut will undercount the saving by roughly one-third on gp2 fleets that were sized this way.

The third cost is tail risk. Small gp2 volumes rely on a burst-credit model for peak IOPS — when credits drain, the volume falls back to its size-based floor, which can be as low as 100 IOPS. This creates latency spikes that get escalated as incidents, investigated, and resolved by resizing the volume (spending more money). gp3 has no credit mechanism; its 3,000 IOPS baseline is always available. Quantifying incident-response cost is harder, but it's real.

Taken together, the true saving opportunity from migrating a gp2 fleet often approaches 35–40% off current EBS spend — not the headline 20% — once right-sizing follows the type migration. That's the number worth surfacing in a business case, and it's the number finance should hold engineering accountable for realizing.

Every AWS account that launched workloads before 2021 likely has some storage still on gp2, the older EBS volume type. AWS replaced it with gp3, which costs about 20% less per GB and delivers better baseline performance. There is no technical reason to stay on gp2; it is purely an inertia issue.

The business case is simple: gp2 volumes are a predictable, ongoing overpayment. The migration is an online, no-downtime API call — no maintenance window, no service disruption. The saving is immediate and recurring. The leadership question is not whether to do this but why it hasn't been done yet, and whether the team has a process for catching these default-setting inefficiencies before they compound over years.

At one company, the VP of Engineering, Tariq, got a report showing $380,000 per year in gp2 storage that had never been migrated. The volumes had been gp2 since launch — nobody had made a deliberate choice to stay on the older type; it just never came up. The migration itself took one engineer one afternoon.

Tariq's response wasn't just to approve the migration. He asked the harder question: how do we make sure the next generation of volumes doesn't default to the wrong type for another three years? The answer was two changes: set gp3 as the default in all infrastructure templates going forward, and add a gp2 volume count to the monthly cloud efficiency report so new drift gets caught within 30 days, not 36 months.

That's the maturity signal for this kind of finding. Closing the current gap is table stakes. The process change — default settings updated, monthly check in place — is what separates a one-time cleanup from a governed state where inertia-driven overspend can't quietly compound.

gp3 has been available since December 2020. If your organization still has gp2 volumes, the reason is inertia — not a deliberate trade-off. Nobody chose to pay 20% more; the volumes were launched before gp3 existed and never revisited. That's a process failure, not a technical one, and it's worth naming clearly.

The business impact is straightforward: every month gp2 volumes remain unconverted, the organization is paying a premium with no return. At fleet scale the annual cost easily reaches six figures. Unlike Reserved Instance commitments or Savings Plans, there is no contractual reason to delay and no technical risk in the migration — the API call is online and takes seconds to issue.

The governance question this surfaces is whether cloud defaults are being reviewed on a cadence. gp2-to-gp3 is the most visible example, but the same pattern — old defaults left in place because no one owns the review — applies to dozens of AWS settings. An organization that closes gp2 quickly and updates its provisioning templates to prevent recurrence has a different maturity level than one that closes it reactively when the finding lands on a report.