Maximize your SQL Server investments on RDS with 5th Gen AMD EPYC Server CPUs!

With the recent adoption of 5th Gen AMD EPYC CPUs by Amazon Relational Database Service (RDS) for SQL Server, AWS customers are expected to provide significant performance and efficiency gains for enterprise database workloads. These 8th generation AWS instances, powered by 5th Gen EPYC processors, provide architectural advancements in processing power and memory bandwidth. This represents a pivotal shift in the enterprise database landscape on AWS. For customers running mission-critical workloads, AMD CPU-powered Amazon EC2 instances offer high performance, cost-effective computing, particularly instances with the 5th Gen AMD EPYC processors. By using these advanced processors with Amazon RDS for SQL Server, a managed service, AWS is not just offering a new CPU option; it is redefining the performance-per-watt standard for relational databases. For AWS customers with little experience with AMD, this adoption is public validation from AWS that AMD infrastructure is production-ready for their most demanding, mission-critical database workloads. Given the historically rigorous data required for a foundational managed service like Amazon RDS for SQL Server to adopt new compute options, this deployment option provides expanded compute options by introducing RDS on instance families, including those powered by 5th Gen AMD EPYC CPUs. This expansion offers significant cost-performance benefits for database workloads

For readers unfamiliar with Amazon RDS for SQL Server, this section provides a brief overview with links to resources for more information. Microsoft SQL Server is a relational database management system developed by Microsoft, beginning with SQL Server 1.0 in 1989. RDS launched support for Microsoft SQL Server in May 2012. With Amazon RDS for SQL Server, you can deploy multiple editions of SQL Server in minutes with cost-efficient and re-sizable compute capacity. Amazon RDS for SQL Server frees you up to focus on application development by managing time-consuming database administration tasks, including provisioning, backups, software patching, monitoring, and hardware scaling.

Amazon RDS for SQL Server supports the “License Included” licensing model for SQL Server. License-included pricing is inclusive of software, underlying hardware resources, and Amazon RDS management capabilities.

Amazon RDS for SQL Server offers hourly pricing with no upfront fees or long-term commitments. In addition, you also have the option to purchase Reserved DB Instances under one-or three-year reservation terms. Reserved DB Instances offer a low, one-time, upfront payment for each DB Instance and then pay a discounted hourly usage rate, up to 69% net cost savings, see here for more details.

On 05/07/2026 Amazon RDS for SQL Server started offering 5th Gen AMD EPYC Server CPUs for deployment in select regions (N. VA, Ohio, Oregon, and more) You can now deploy db.m8a.medium-48xl (including db.m8a.metal-24xl-48xl) and db.r8a.medium-48xl (including db.r8a.metal-24xl-48xl) instances. The rest of this post will focus on the benefits 5th Gen AMD EPYC processors bring to Amazon RDS for SQL Server customers.

AMD powered instances used by Amazon RDS for SQL Server

The Amazon Elastic Compute Cloud (EC2) instances selected by the Amazon RDS for SQL Server service are m8a.medium-48xl (including m8a.metal-24xl-48xl) and r8a.medium-48xl (including r8a.metal-24xl-48xl) instances. The same model of 5th Gen AMD EPYC (9R45) processor powers the instances, such as c8a and hpc8a.

CPU and Amazon EC2 specifications

Before we dive into the AWS specifics, let’s look under the hood of the 5th Gen AMD EPYC CPU. The AMD EPYC CPU is the fifth generation of EPYC server-class CPUs released by AMD since 2017. It is based on the “Zen 5” architecture and built using 4nm CPU process technology. AMD designed the EPYC CPU on AWS with a chiplet architecture. In traditional monolithic architecture, all the CPU cores on a chip share memory bandwidth and L3 cache. This can lead to a “noisy neighbor” situation where a single instance’s vCPUs could consume most L3 cache and memory bandwidth which will constrain those resources for other instances to consume. With AMD chiplet architecture, each 8-core CCD (chiplet) has its own dedicated L3 cache and memory bandwidth, in a noisy neighbor situation limits the blast radius for resource contention to the cores that comprise the chiplet instead of all cores in the CPU.

AMD EPYC 9R45 CPU detailed specifications

• Amazon EC2 Instance families: c8a, hpc8a, m8a (excl. m8azn), r8a
• Sockets per host: 2
• Core Architecture: “Zen 5”
• CPU Process Technology: 4nm
• I/O Die Process Technology: 6nm
• Cores: 96
• Cores per CCD: 8
• L3 Cache: 384 MB
• Max Frequency: 4.5GHz
• Memory channels: 12 
• Memory type: DDR5-6400
• Max Power Consumption: 320W (hpc8a = 500W)
• CPU Options:
  • Advanced Vector Extensions (AVX) 3-512 (single 512-byte pipe vs two 256-byte pipes in 7a / Genoa)
  • AVX-512 Vector Neural Network Instructions (VNNI)
  • Bfloat16
  • AMD Secure Memory Encryption (AMD SME)
  • AMD Secure Encrypted Virtualization - Secure Nested Paging, disabled
  • AMD Secure Encrypted Virtualization - Trusted I/O, enabled
  • Simultaneous Multi-Threading (SMT) disabled, every vCPU tied to pCore

The Amazon EC2 instance specifications

The instance families selected by the Amazon RDS for SQL Server service are m8a and r8a. Both instance families are powered by the same 5th Gen EPYC processor model. Their “t-shirt sizes” differentiate the instances; the m8a family offers a 1vCPU to 4GiB RAM ratio, and the r8a family offers instances with a 1vCPU to 8GiB RAM ratio. Deciding between the two is finding the size that matches the requirements at the desired cost.

All 8a instance families are built on AWS’s Nitro system. Specifically, the 8a instance families are built on the latest generation Nitro v6. The Nitro System is a collection of hardware (Nitro cards) and software components built by AWS that enable high performance, high availability, and high security. This allows AWS to get the maximum value and performance out of its hosts.

Performance of Amazon RDS for SQL Server with AMD

Now that you have learned about what is powering Amazon RDS for SQL Server, let’s dive into the benchmarks that back up the performance and efficiency claims. AMD benchmarked a variety of “m” and “r” Amazon EC2 instance types to show AMD performance and cost advantages to Amazon RDS and its customers.

The benchmark setup

For benchmarking purposes, AMD utilized commercially available Amazon EC2 instances without implementing specific modifications intended to enhance AMD performance or diminish Intel's results. We used an OLTP workload derived from TPC-E to load the instances, and the details are described below. This benchmark is derived from a TPC benchmark and is not comparable to published TPC results. Keep in mind these benchmarks only look at CPU efficiency and performance. IO and memory performance were not evaluated because the AWS platform controls and defines those parameters.

• Microsoft Windows Server 2022 w/ SQL Server 2022 CU1 Amazon EC2 instances tested:
  • r8a vs. r8i: xl, 2xl, 4xl, 8xl, 16xl
• EBS storage configuration:
  • Logs disk: (1) io2, 5K IOPS, 1000GiB
  • DB disk: (1) io2, 80K IOPS, 399GiB
• Microsoft TPC benchmarking tools generated TPC-E load.
  • OLTP workload derived from TPC-E

The results:

Diving into the results, you will see AMD holds sizable transactions per second (TPS) performance and cost per TPS advantage over its Intel-powered counterparts. Keep in mind that AMD estimates that Intel-powered Amazon EC2 instances consume over 585 watts, where the AMD CPU-powered instance only consumes 320 watts. See here for more on how AMD calculated the power consumption of Amazon EC2 instances. Also, keep in mind these performance results are from regular Amazon EC2 instances, not from Amazon RDS for SQL Server. The customer value for Amazon RDS for SQL Server will also apply to running Microsoft SQL Server on Amazon EC2 instances.

R8a versus R8i

The r8a and r8i instances represent the latest generation of server-class CPUs from Intel and AMD powering AWS’s mainstream memory-optimized instances.

The first chart shows the Microsoft SQL Server transactions with price performance per second normalized to the performance of the Intel-powered instances Excluding SQL Server license cost, is a simple statistic where the CPU processing more DB transactions per second is the performance leader. The AMD CPU-powered instances’ transactions per second are up to 63% better compared against the same size / generation Intel-powered instances. 

Now that performance has been demonstrated, what does this mean for licensing? With Amazon RDS for SQL Server, you are billed per vCPU for the Windows and SQL Servers licenses applied to the overall Amazon RDS DB Instance cost. Fewer vCPUs lead to reduced instance costs due to the reduced licensing burden. Windows and MSSQL licensing costs combined are over 80% of an instance’s costs.

In the chart below, we have normalized the cost per transaction per second to the Intel-powered instances including SQL server licensing. The data shows AMD is anywhere from $0.37 per transaction per second to $0.23 cheaper than its Intel peer instance. The performance data above, while a great story, only shows the raw performance of the CPU. The actual cost of processing the workload is where business decisions are made. AWS customers using AMD-powered instances with Amazon RDS for SQL Server achieve industry-leading cost-efficient transactions per second compared to any other x86 CPU on Amazon RDS for SQL Server.

Summary

The data is clear: AMD holds a sizable performance and total cost advantage for Microsoft SQL Server workloads on AWS. Consider Amazon RDS customers can now optimize their Amazon RDS instance’s vCPU count, which can drive further savings on their Amazon RDS for SQL Server deployments. We will post a follow-up blog post detailing how you can properly size your Amazon RDS for SQL Server and self-hosted SQL Server Amazon EC2 instances to drive additional value from your AWS investments.

Don’t Go At It Alone

Whether you’re modernizing, downsizing, or simply looking for cost efficiency for your SQL Server workloads on AWS in Amazon RDS or on Amazon EC2, AMD offers expertise to guide the way. Many organizations combine multiple strategies across different workloads to achieve maximum ROI. If you need help deciding what best fits your architecture, AMD and its partners are ready to assist. To offer feedback on this post or suggest future topics, please reach out to AWS@AMD.com.