For Aurora RDS, how many vCPUs to 1 ACU?

In Aurora Serverless v2, capacity is measured in Aurora Capacity Units (ACUs) rather than vCPUs. Each ACU provides approximately 2 GiB of memory with corresponding CPU and networking resources. However, AWS doesn't provide a direct 1:1 conversion between ACUs and vCPUs.

When planning to convert from Aurora Serverless v2 to provisioned instances, you'll need to consider your memory requirements first, since that's how ACUs are primarily defined. For example, if you're using 8 ACUs in Serverless v2, that would be equivalent to approximately 16 GiB of memory.

To find the right provisioned instance type, you should:

1. Determine your current ACU usage (minimum, maximum, and average)
2. Calculate the equivalent memory (ACUs × 2 GiB)
3. Select a provisioned instance type that provides similar memory capacity

For instance, if your Aurora Serverless v2 database is using 16 ACUs (32 GiB of memory), you might consider a db.r6g.xlarge or similar instance that provides comparable memory.

For the most accurate comparison, you can monitor your actual ACU usage in CloudWatch metrics before making the conversion decision.
Sources
Evaluating the right fit for your Amazon Aurora workloads: provisioned or Serverless v2 | AWS Database Blog
Moving from Aurora to Aurora Serverless | AWS re:Post

There is no official documentation from AWS that provides a direct conversion ratio between Aurora Capacity Units (ACUs) and vCPUs. This is because an ACU is a composite unit that includes not only CPU, but also memory, networking, and I/O.

However, based on user-provided information and community discussions, a common approximation is that 1 ACU provides approximately 2 GiB of memory, with the corresponding CPU and networking.

Here's a breakdown of what that means for your conversion:

• ACU and Memory: The most consistent and documented part of an ACU is its memory component: 1 ACU equals approximately 2 GiB of RAM.
• ACU and vCPU: The relationship between ACUs and vCPUs is not a fixed ratio. The "corresponding CPU" is dynamic and depends on the specific instance type and workload. While some users have offered rough estimates (e.g., 1 ACU might correspond to 0.25 vCPU or a certain vCPU-to-memory ratio), these are approximations and not official AWS figures.
• Pricing: When you are calculating costs, it's important to know that Aurora Serverless v2 is priced per ACU-hour, not per vCPU-hour. This is a key difference from provisioned instances, which are billed based on instance type (e.g., db.t4g.medium with a fixed number of vCPUs and memory).

How to Estimate Costs for Provisioned Instances

Since you are moving from Serverless to provisioned, the best way to calculate costs is to find the provisioned instance type that matches your memory requirements. Since 1 ACU is about 2 GiB of RAM, you can estimate your provisioned instance size based on your typical or peak ACU usage.

For example, if your Aurora Serverless instance typically runs at 8 ACUs, that's roughly 16 GiB of memory. You would then look for a provisioned Aurora instance class with a similar amount of memory, such as an r6g.large or r5.large (which typically have 2 vCPUs and 16 GiB of RAM), to get a cost estimate. .

Keep in mind that this is still an estimation, and your actual costs will vary based on your workload and other factors like I/O usage, backup storage, and data transfer.