Aurora vs RDS: how they differ on cost, performance and limits
How Amazon Aurora and Amazon RDS differ in storage, I/O billing, replicas, limits and price, read from AWS's own pages and price list on 17 Nov 2026.
45 minutes with an engineer: we review your workload and sketch the design.
- ✓OSStend is our service
- ✓Checked 17 Nov 2026
- ✓US East (N. Virginia)
Aurora separates compute from a shared, multi-zone storage volume and bills storage as used plus I/O requests, while RDS runs the engine on provisioned EBS storage. RDS is simpler and usually cheaper at steady, modest I/O; Aurora adds up to 15 read replicas and 256 TiB of storage. Checked 17 Nov 2026, US East.
- PostgreSQL
- MySQL
- MongoDB
- Redis
- Valkey
- OpenSearch
- Kafka
- RabbitMQ
- AWS
- Google Cloud
- Azure
- Kubernetes
- OpenShift
- Prometheus
- Grafana
- Datadog
- Dynatrace
- Splunk
Aurora vs RDS in one table
Aurora is built for read scaling and fast replica failover inside AWS and bills storage by use plus I/O; RDS is the plain managed engine on provisioned storage with a lower instance rate. The table sets them side by side, with the source and date beside every fact.
| Amazon Aurora (PostgreSQL or MySQL) | Amazon RDS (PostgreSQL or MySQL) | |
|---|---|---|
| Storage model | A cluster volume spanning multiple Availability Zones, each holding a copy; you pay for one copy (AWS docs, pricing, checked 17 Nov 2026) | EBS volumes you provision per instance (AWS docs, checked 17 Nov 2026) |
| Storage billing | $0.10 per GB-month consumed (Standard) or $0.225 (I/O-Optimized) | $0.115 per GB-month provisioned, gp3; $0.23 Multi-AZ |
| I/O billing | $0.20 per million requests on Standard; none on I/O-Optimized | No per-request charge; IOPS above the baseline are billed per IOPS-month |
| Replicas | Up to 15 Aurora Replicas on the same volume, usually with replica lag well under 100 ms per AWS (docs, checked 17 Nov 2026) | Read replicas billed as separate DB instances with their own storage (docs, read 15 Nov 2026); Multi-AZ DB clusters add two readable standbys (docs, checked 17 Nov 2026) |
| High availability | Failover to a replica in another zone; replicas are billed as instances | Multi-AZ instance: one standby that cannot serve reads (docs, checked 17 Nov 2026) |
| Maximum storage | Up to 256 TiB for compatible versions (docs, checked 17 Nov 2026) | 64 TiB for PostgreSQL and MySQL instances (docs, checked 17 Nov 2026) |
| Instance classes | r-class (memory) and t-class (burstable) families; no m-class in the price list | m, r and t classes |
| Serverless option | Serverless v2: $0.12 per ACU-hour (Standard), $0.16 (I/O-Optimized) | None |
| Backup storage | Free up to 100% of cluster size; $0.021 per GB-month beyond | Free allocation, then $0.095 per GB-month |
| Where it runs | AWS only | AWS only |
How we compared: AWS's Aurora and RDS pricing pages, user-guide pages and the AWS Price List file for Amazon RDS in us-east-1 (version 20261113215233, published 13 Nov 2026), on-demand USD, checked 17 Nov 2026. We did not run benchmarks and quote none. OSStend is our service and runs neither, so read the verdicts below as one vendor's view.
How Aurora and RDS differ in architecture
Aurora puts the data on a shared storage volume that every instance in the cluster reads, so adding a reader adds compute but no new copy of the data. RDS gives each instance its own EBS volume, and a standby or replica holds its own copy.
AWS describes an Aurora cluster as a primary writer plus up to 15 Aurora Replicas attached to one cluster volume that keeps a copy in each Availability Zone, and says an Aurora instance is still a cluster because the storage is spread across zones (AWS docs, checked 17 Nov 2026). RDS offers a Multi-AZ instance with a non-readable standby, or a Multi-AZ DB cluster with two readable standbys and semisynchronous replication (AWS docs, checked 17 Nov 2026). AWS says Aurora updates require a restart, which typically means 20 to 30 seconds of downtime (AWS docs, checked 17 Nov 2026).
Aurora vs RDS cost on the same size
On db.r7g.xlarge (4 vCPU, 32 GiB) with 500 GB, an RDS Multi-AZ primary lists at about $813 a month and an Aurora writer plus one reader at about $857 before I/O; Aurora I/O-Optimized lists at about $1,162. The break-even for I/O-Optimized on this shape is about 1.5 billion I/O requests a month.
| Shape, 500 GB | Arithmetic, list prices | Monthly total |
|---|---|---|
| RDS PostgreSQL, one node, Single-AZ | $0.478 x 730 = $348.94, plus 500 x $0.115 = $57.50 | $406.44 |
| RDS PostgreSQL, Multi-AZ standby | $0.956 x 730 = $697.88, plus 500 x $0.23 = $115.00 | $812.88 |
| Aurora PostgreSQL Standard, writer only | $0.553 x 730 = $403.69, plus 500 x $0.10 = $50.00, plus I/O | $453.69 plus I/O |
| Aurora PostgreSQL Standard, writer plus one reader | 2 x $403.69 = $807.38, plus 500 x $0.10 = $50.00, plus I/O | $857.38 plus I/O |
| Aurora PostgreSQL I/O-Optimized, writer plus one reader | 2 x ($0.719 x 730 = $524.87) = $1,049.74, plus 500 x $0.225 = $112.50 | $1,162.24 |
Labelled arithmetic on US East on-demand list prices, db.r7g.xlarge, 730-hour month, checked 17 Nov 2026, from the AWS Price List file. I/O on Standard is $0.20 per million requests: 500 million requests adds $100 and 5 billion adds $1,000. Break-even between Standard and I/O-Optimized on this shape: ($1,162.24 - $857.38) / $0.20 per million = about 1,524 million requests a month. Excludes transfer, extra backup storage and taxes. See the cost breakdown for RDS.
Two things shift these totals: how many I/O requests your workload makes, which you can read from CloudWatch before you decide, and whether you need readers at all. Aurora's I/O-Optimized option exists for I/O-heavy workloads, and AWS says it can save up to 40% when I/O exceeds 25% of Aurora spend.
From our on-call engineers: how we decide between Aurora Standard and I/O-Optimized
The Aurora decision that moves the bill most is the storage configuration, not the instance size. AWS's own rule of thumb is that I/O-Optimized wins when I/O is 25% or more of total Aurora spend, so we measure a full month of I/O, including batch jobs, before choosing.
- ✓AWS says Aurora I/O-Optimized is the better choice when I/O spend is 25% or more of total Aurora database spend, and Standard when it is below 25%.
- ✓You can move from Standard to I/O-Optimized only once every 30 days, so we do not flip the setting during a month we have not measured.
- ✓Switching on NVMe-based instance classes needs an engine restart, while non-NVMe instances switch with no downtime, so the maintenance window depends on the class.
- ✓Aurora keeps copies of the data across three Availability Zones in a cluster volume independent of instance count, which is why adding a reader does not copy table data.
- ✓Aurora bills only for storage in use and shrinks the allocation when you drop data, so dropping unused tables and indexes reduces the bill.
What to test yourself instead of trusting performance claims
Neither AWS page gives a single throughput number that holds for your workload, so test your own queries on both. We quote no benchmark: the parts worth measuring are write latency with a standby, replica lag under your write rate, failover time and I/O request volume.
- Write latency.
- AWS notes Multi-AZ instance deployments can have higher write and commit latency than Single-AZ because replication is synchronous (AWS docs, checked 17 Nov 2026).
- Replica lag.
- For RDS Multi-AZ DB clusters, AWS says failover time depends on replica lag (AWS docs, checked 17 Nov 2026). For Aurora, measure lag at your write rate.
- I/O volume.
- Aurora Standard bills per request, so read the I/O metrics for a week before you price it.
- Failover.
- Trigger a failover in a test environment and time it with your application's connection handling.
Aurora and RDS limits that matter
The limits that decide most choices are storage ceilings, the replica count, instance families and where it can run: Aurora reaches 256 TiB and 15 replicas, RDS instances reach 64 TiB, and both exist only on AWS.
- Storage.
- Aurora up to 256 TiB for compatible versions, RDS PostgreSQL and MySQL up to 64 TiB (AWS docs, checked 17 Nov 2026).
- Replicas.
- Aurora allows up to 15 Aurora Replicas; RDS read replicas are separate instances.
- Engines.
- Both offer PostgreSQL- and MySQL-compatible engines; MariaDB is an RDS engine and does not appear as an Aurora engine in the price list read.
- Cloud.
- Both run only on AWS infrastructure, so the same design cannot move to another cloud or a data center.
When to pick Aurora, when RDS, and when neither
Pick RDS for a standard workload with modest I/O and a standby; pick Aurora when you need many read replicas, very large storage or fast scaling of readers; pick neither when the database must live outside AWS or needs settings neither offers.
Contact us if
- ✓RDS: a standard PostgreSQL or MySQL workload on AWS with a Multi-AZ standby and steady I/O.
- ✓Aurora: read-heavy workloads that need several readers on one volume, or data beyond 64 TiB.
- ✓Aurora I/O-Optimized: an I/O-heavy workload where I/O is over a quarter of the Aurora bill.
Not a fit if
- ✕The database must run in another cloud, in your own account, or in a data center.
- ✕You need extensions, versions or tuning outside what AWS offers.
- ✕You want one design and one price model across clouds rather than per-service AWS lines.
OSStend fits the second list: dedicated PostgreSQL and MySQL designed to the workload and run 24/7. See Amazon RDS alternatives, managed PostgreSQL, or price your own shape in the cost calculator.
How we compared
Every fact comes from AWS's own pricing pages, user guide and price list, read on 17 Nov 2026, with the link beside it. No benchmarks were run, and the cost examples are arithmetic on list prices.
We compared like with like where AWS allows it: db.r7g.xlarge, because Aurora's price list has no m-class instances, in US East, on demand. OSStend runs neither service, but it sells dedicated databases, so it has an interest in the answer.
FAQ
Is Aurora faster than RDS?
AWS pages do not give one number that holds for every workload, and this page quotes no benchmark. Test your own queries, with and without a standby, and measure replica lag and failover time.
Is Aurora more expensive than RDS?
Per instance hour, yes: db.r7g.xlarge lists at $0.553 on Aurora PostgreSQL and $0.478 on RDS PostgreSQL (checked 17 Nov 2026). Storage and I/O billing differ, so the total depends on your I/O volume and replicas.
What is Aurora I/O-Optimized?
A pricing mode with no per-request I/O charge, higher instance rates and $0.225 per GB-month storage. AWS says it can save up to 40% when I/O is over 25% of Aurora spend (Aurora pricing, checked 17 Nov 2026).
What is the difference between Aurora PostgreSQL and RDS for PostgreSQL?
Aurora PostgreSQL is PostgreSQL-compatible software on Aurora's shared storage volume; RDS for PostgreSQL runs the community engine on EBS storage per instance. See the table above for billing, replicas and limits.
Can I run Aurora outside AWS?
No. Aurora and RDS run on AWS only. A dedicated PostgreSQL or MySQL cluster can run in another cloud, in your own account or in a data center; see OSStend managed PostgreSQL.
Related pages
- ComparisonAWS RDS pricing explained (2026)AWS RDS and Aurora pricing from AWS list prices: instances, storage, I/O, backups and Multi-AZ, with a cost calculator and when RDS is the cheaper choice.
- AlternativeAmazon RDS alternatives in 2026Amazon RDS alternatives for PostgreSQL and MySQL: where RDS is the better choice, what the alternatives cost, and running dedicated clusters in any cloud.
- PostgreSQLManaged PostgreSQL service and Postgres hostingManaged PostgreSQL service, compared: how to choose between RDS, Aiven, Crunchy Bridge, Neon and a tailored cluster run 24/7 in your cloud or ours.
- Tools · CostManaged database cost calculatorFree managed database cost calculator: price one cluster self-run, on a cloud database service and with managed providers, from sourced, dated list prices and your team's time.
Bring your AWS database bill to an engineer
Share the engine, size, replicas and I/O. An engineer will say whether Aurora, RDS or a designed cluster is the right fit.
- ✓A design sketch for your workload, not a plan menu
- ✓A zero-downtime cutover path for what you run today
- ✓A price for that design, after the review