The Same Total io2 IOPS Can Have a Different IOPS Charge

Price the provisioned rate on each volume

AWS's io2 pricing explanation, checked October 21, 2026, separates provisioned storage from provisioned IOPS and explicitly applies IOPS tiering per volume. Its first tier covers the first 32,000 IOPS; the second covers the quantity from 32,001 through 64,000 IOPS. This comparison stays within those two tiers.

A tier boundary divides the provisioned quantity. A volume at 32,000 IOPS has no second-tier quantity. A volume at 32,001 has one IOPS in that tier, while its first 32,000 remain in the first tier. Crossing the threshold does not reprice every IOPS on the volume at the second-tier rate.

The priced quantity is a provisioned performance rate held over time. It is not a count of completed application operations. Multiplying that rate by elapsed seconds to estimate operations would answer a different question from calculating an IOPS-month charge.

References: [AWS-P]

IO2-T48: one 48,000-IOPS volume versus two at 24,000

Consider two hypothetical offers using the same applicable io2 price schedule for the same full monthly pricing interval. Offer A provisions 48,000 IOPS on one volume. Offer B provisions 24,000 IOPS on each of two volumes. Both totals are 48,000 provisioned IOPS throughout that interval. This is a charge-component illustration, with no specified capacity, instance or approved configuration.

Let r1 be the first-tier rate and r2 the second-tier rate, each measured in monetary units per provisioned-IOPS-month. They are symbolic inputs, with no numerical dollar rate or regional quote supplied. Because the quantities remain provisioned for the full pricing interval, each monthly rate applies for one full interval.

For A, the first-tier quantity is 32,000 IOPS. The remaining 48,000 - 32,000 = 16,000 IOPS enter the second tier. Its provisioned-IOPS component is therefore A = 32,000r1 + 16,000r2.

For B, each 24,000-IOPS volume stays below the first threshold. Each contributes 24,000r1, so B = 24,000r1 + 24,000r1 = 48,000r1. B's first-tier quantities sum to 48,000, but neither individual volume has a second-tier quantity.

Subtracting gives B - A = 16,000(r1 - r2). If the applicable r1 exceeds r2, B has the higher provisioned-IOPS component. If the rates are equal, the difference is zero. Exactly 16,000 IOPS move between the price tiers in this comparison; the total purchased IOPS rate remains unchanged.

References: [AWS-P]

Pooling first erases the difference

An incorrect calculation might add B's two quantities first, obtain 48,000 IOPS, and apply the tiers once to that total. It would assign B 32,000 first-tier IOPS and 16,000 second-tier IOPS, returning the same expression as A.

That calculation invents a second-tier quantity for B. Neither of its volumes crosses 32,000 IOPS. It understates B's component by 16,000(r1 - r2) when r1 exceeds r2, exactly concealing the difference the comparison is meant to expose.

The opposite shortcut also fails: charging all of A's 48,000 IOPS at r2 discards the first-tier quantity that remains on that volume. The correct order is to assign each volume's quantities to the applicable tiers, calculate its component for its interval, and then sum the components. An aggregate IOPS headline cannot replace those intermediate quantities.

References: [AWS-P]

Retain each volume's billable interval

AWS describes io2 provisioned storage and IOPS billing in per-second increments with a 60-second minimum. A monthly rate unit therefore does not imply that every volume is purchased for a full month. IO2-T48 deliberately fixes a common full interval so that timing does not obscure the tier effect.

For unchanged quantities and rates over a shorter interval, weight each volume's tiered monthly component by its own billable fraction of the pricing month, applying the actual billing increments and minimum. If all volumes share the same fraction f, the illustrated difference becomes f x 16,000(r1 - r2). That common factor changes the amount, while leaving the per-volume tier assignment intact.

If the intervals differ, retain their separate weights before summing. A shorter interval on one volume does not move another volume's IOPS into a different tier. If provisioned quantities or applicable rates change, preserve the relevant intervals rather than treating an opening, closing or average aggregate IOPS figure as the whole purchase.

Use the related retained-resource billing article when the question is which resources remain chargeable during a compute pause. Here, the necessary input is the billable interval for each provisioned quantity, rather than the GPU job's running time.

References: [AWS-P]

Assess whether the layouts can serve the same workload

The arithmetic does not establish that either offer can be provisioned for the intended environment. AWS's io2 documentation ties provisioned performance to volume and instance support conditions. The actual capacity, provisioned IOPS and supported attachment environment belong in the platform owner's assessment.

Nor does 48,000 total provisioned IOPS establish equal attainable performance. AWS bounds EBS performance by the smaller of the instance's limits and the attached volumes' aggregate performance; IOPS and byte-throughput limits also interact with I/O size. The related article More Provisioned Volume Performance Can Leave the Same EBS Bottleneck develops that comparison.

Work distribution matters to substitutability. Two volumes with 24,000 IOPS each do not establish that work directed to one of them can use the other's allowance. Conversely, a single volume's lower IOPS component does not show that datasets or owners may share the same storage boundary.

Ask the responsible platform and application owners whether the required workload distribution, isolation and availability arrangements permit either layout. A lower component is useful purchase evidence only alongside a feasible layout that meets those requirements. IO2-T48 supplies no recommendation to merge volumes.

References: [AWS-E][AWS-V]

Request a reproducible component comparison

Ask the proposer to identify each offered volume, its provisioned IOPS, its first- and second-tier quantities where applicable, and the rates and billable interval used for that line. The applicable AWS Region, currency and pricing date connect the symbolic rates to the actual offer. A response containing only 48,000 total IOPS leaves this comparison unresolved.

Keep provisioned capacity and its charge separate, then include other actual resources and applicable commercial adjustments when constructing the complete bill. Equal IOPS totals do not establish equal provisioned capacity, equal capacity charges or equal overall cost. The difference derived here concerns only the stated provisioned-IOPS component.

A later allocation of those resource charges among internal teams is another decision. Changing a team split cannot correct an incorrectly calculated provider component. The related shared-platform allocation article explains that subsequent accounting question.

References: [AWS-P]

Scope a separate physical cabling need

If the wider program also includes connections for customer equipment, Leviathan's Structured Cabling & Fiber Installation service covers agreed physical routing, installation, labeling and connection records using customer-furnished or separately sourced materials.

Use the service and project-inquiry links below with a nonconfidential equipment and site summary, approved endpoints, material responsibilities and required handoff records. That inquiry concerns the separate physical installation scope. The appointed cloud and application owners retain the io2 pricing and storage-layout decisions.

References: [L1]