Testing Your Cloud's Power Efficiency

In most cloud environments, customers don't have to think about how much power their data centers use. Still, in some regions, organizations are responsible for reporting carbon emissions tied to their cloud workloads. Since the power usage of an Amazon Elastic Compute Cloud (EC2) instance directly affects its environmental footprint, improving efficiency helps both your organization and the planet. It isn’t just a theoretical concern: in cities like Dublin and Frankfurt, new data centers are facing delays of up to three years just to connect to the power grid due to capacity limits.

Here’s the good news: our testing shows that AMD-powered Amazon EC2 instances are not only more power-efficient but also deliver 2.7x better performance per watt compared to Intel-based instances. By consuming 213 fewer watts on average, AMD instances offer a clear advantage for organizations looking to reduce their environmental impact and optimize costs.

Let’s dive into how we conducted these tests, the tools we used, and how you can measure the power consumption of AMD-powered Amazon EC2 instances using bare-metal configurations.

How We Tested Power Consumption

To measure the hardware power usage of your instances accurately, you’ll need to use a bare-metal Amazon EC2 instance. That’s because the model-specific register (MSR) data required for power readings is only available on bare metal. With bare-metal instances, your operating system talks directly to the physical CPU with no hypervisor in the way. On standard EC2 instances, the hypervisor blocks MSR access, and even if it didn’t, CPU scheduling would make the results unreliable.

It’s also important to note that CPUs pull their maximum power only when they’re under load. Modern processors dynamically adjust power draw, so to see true peak consumption, you must stress the CPU while collecting data.

To simplify the testing process, we automate everything from end to end. When you deploy and run the provided AWS Systems Manager (SSM) Automation document via the AWSvCloudFormation template, it sets up the environment, runs the tests, sends the metrics to Amazon CloudWatch, and then cleans up all the resources afterward.

What the Automation Deploys

The SSM Automation workflow spins up the following resources:

• 1 Amazon Virtual Private Cloud (VPC)
• 1 Amazon VPC Internet Gateway (IGW), attached to the VPC
• 1 Public VPC Subnet
• 1 custom VPC route table with a 0.0.0.0/0 route through the Internet Gateway
• 1 AWS Identity and Access Management (IAM) Amazon EC2 instance role with AmazonSSMManagedInstanceCore, CloudWatchAgentServerPolicy, and a custom inline policy
• 2 Amazon EC2 Metal Instances:
  • AMD-powered: m8a.metal-48xl
  • Intel-powered: m8i.metal-48xl

Instance Configuration

All tests are rerun on Ubuntu Server 24.04 (/aws/service/canonical/ubuntu/server/24.04/stable/current/amd64/hvm/ebs-gp3/ami-id). We initially tried Amazon Linux 2 but had issues retrieving power data. You may have different results depending on your Linux distribution.

These instances use their default settings, with only a few required additions.

The Tools to Install

• System tools: unzip, curl, Python 3, linux-tools-common, linux-tools-generic, msr-tools, stress-ng, AWS CLI.
• Turbostat collects CPU metrics such as frequency, idle states, temperature, and, especially, PkgWatt, which we use to measure power draw.
• Stress-ng, which applies CPU load and reports “bogus operations per second (bogos)” so we can measure consumption under maximum stress.

Deploying and Running the Test Environment

We’ve designed the testing process to be fully automated. To get started, simply download and deploy the CloudFormation template, which provisions the Power Consumption SSM Automation document. If you're new to CloudFormation, setup instructions are available in AWS documentation.

Once the SSM Automation document is deployed, you can run it manually by following the same documentation. After you invoke it, the workflow handles everything and removes all resources once the tests are complete.

One important note: Do not run these tests in a production AWS account. If you'd like to try them yourself, use a non-production or sandbox AWS environment.

What to Expect When Running the Test

The test process takes up to 30 minutes to complete. There is a 10-minute sleep timer to ensure the instances terminate before removing all other components. After a test run is complete, you can go to CloudWatch to review the results.

1. Open the CloudWatch console.
2. On the left-hand navigation pane under Metrics, select All metrics.
3. In the bottom pane of the metrics console, select the Custom namespace titled CPUPowerTest.

4. Within the Custom namespace, select the dimension InstanceID.

5. There you will see a list of instances that have published metrics to CloudWatch. Select the instance(s) you would like to review. If you have a large number of instances, you will need to know the InstanceID of the instances you want to review.

Note: Each instance will have two metrics: PowerConsumptionWatts (amount of wattage the CPU consumed) and PerformancePerWatt (Bogo Ops per second/wattage). 

6. To make the data easier to consume, we recommend you use the "Data table" view to review the information

The following two graphs plot all these results. The first one shows power consumption, and the second one shows performance per watt. 

Conclusion: AMD has the Power Efficiency Advantage

AMD-powered Amazon EC2 instances are the most power-efficient and most powerful x86 processors on AWS. Our testing revealed that, at maximum power, AMD instances draw 196 fewer watts than their Intel counterparts and, on average, they consume 213 fewer watts. Despite using significantly less power, AMD outperforms Intel by an impressive 2.7x (or ~175%) in performance per watt.

This isn’t just about numbers. The results highlight broader implications for cloud users and the environment. For organizations in regions with strict carbon reporting or power grid constraints, AMD-powered instances provide a clear path to reducing environmental impact while maintaining high performance. Lower power consumption also translates to cost savings over time.

It’s important to note that this test wasn’t designed as a benchmark but as a practical validation of system performance under real-world conditions. The results provide insight into the power demands faced by hyperscalers like AWS and highlight the critical role efficient processors play in meeting those demands sustainably.

For businesses modernizing their cloud workloads, AMD-powered EC2 instances offer a compelling choice for cost, performance, and sustainability. With tools like the AMD EPYC™ Advisory Suite and partnerships with AWS, AMD is ready to help you find the right solution. Try the test yourself or contact us at AWS@AMD.com to learn more.

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 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 to suggest future topics, feel free to reach out at AWS@AMD.com.