Boost Performance and Cut Costs by Lowering the vCPU Count on AMD-Powered EC2 Instances

When you think of improving cloud performance, your first instinct may often be to scale up, adding more virtual CPUs (vCPUs), memory, and storage. Doing so would certainly seem to make sense. But what if scaling down could actually improve performance—and save money?

That's exactly what AMD's latest benchmarking with AWS EC2 instances shows. By using newer AMD EPYC™ Server CPUs in EC2 instances, you can potentially reduce the number of vCPUs in play and still gain performance, especially for database workloads like PostgreSQL.

The Newest AMD-Powered EC2 Instances

Amazon EC2 offers several instance families powered by the latest AMD EPYC 9004 Series processors, including:

• M7a
• C7a
• R7a
• Hpc7a

These 4th Gen AMD EPYC CPUs use DDR5 memory, providing 2.25x the memory bandwidth and up to 50% higher performance compared to the previous AMD 6a instances. AMD has also optimized these processors for better energy efficiency and performance across a wide range of workloads, including database processing and natural language processing (NLP), delivering up to 2.7x the performance-per-watt.

Benchmarking Overview: PostgreSQL on EC2

Let's walk through a real-world PostgreSQL performance test using EC2 instances to see how modernizing and downsizing can be a powerful optimization strategy.

What Was Tested?

AMD benchmarked PostgreSQL 17.5 using the TPC-C workload (a standard for measuring online transaction processing performance). The goal was to prove that downsizing with AMD-powered instances doesn't mean sacrificing performance.

The Test Setup

The benchmark used two EC2 instances:

Note that the AMD-powered r7a.4xlarge has half the vCPUs of the Intel-based r7i.8xlarge.

Other Configuration Details:

• OS: Amazon Linux 2023 (kernel 6.1)
• DB Engine: PostgreSQL 17.5
• Benchmark Tool: HammerDB 4.12
• Disks:
  • OS: 1 x 200GiB io1 (10,000 IOPS)
  • DB Data: 3 x 500GiB io1 (20,000 or 25,000 IOPS)
• File System: ext4
• Setup Consistency: Both instances used the same AMI, database settings, disks, and configurations. HammerDB was run from a separate VM in the same region.

The Results: Performance With Fewer vCPUs

Let's break it down by IOPS configuration.

Test 1: 20,000 IOPS

The key metric here is New Orders Per Minute (NOPM), a standard TPC-C metric for transactional throughput.

Even with half the cores, the AMD r7a.4xlarge outperformed the larger Intel® instance by up to 41%.

Test 2: 25,000 IOPS

The same setup, with higher disk throughput.

The performance delta remained strong with up to 38% better performance for the smaller, AMD-based instance.

The Cost Savings Really Add Up

Let's assume these instances run 24/7 for a year. Here's what the compute costs look like for PostgreSQL on Linux:

You achieve annual savings of $7,880 per instance just by choosing a better-optimized AMD-powered instance.

Now let's look at the cost impact for SQL Server on Windows:

Windows + SQL Server Enterprise

That's a savings of $66,887 per year (per instance).

Windows + SQL Server Standard

That's a savings of $31,147 per year.

The Takeaway: Get Great Performance and Impressive Savings with Fewer Cores

This benchmark test confirms what AMD has been saying all along: you don't always need more vCPUs to get better performance. With the latest generation of AMD EPYC-powered EC2 instances, you can often do more with less.

• Get higher throughput even at lower core counts
• Achieve significant cost savings, especially with SQL Server on Windows
• Find a simplified optimization path by modernizing and downsizing

And the best part? Switching EC2 instance types is a straightforward process that can be automated. While database workloads may need graceful shutdown handling, this change can usually be made during your regular maintenance window.

Don't Go It Alone

Whether you're modernizing, downsizing, or simply looking for cost efficiency, AMD offers tools and expertise to guide the way. The AMD EPYC Advisory Suite can help you identify the right EC2 instance for your workload, and AMD works closely with AWS partners to deliver tailored solutions. Many organizations combine several strategies across different workloads for maximum ROI.

If you need help deciding what fits your architecture best, AMD and its partners are ready to assist. To offer feedback on this post or to suggest future topics, feel free to reach out at AWS@AMD.com.