Migrating from AMD to ARM in AWS EKS. Benefits of ARMs.
I'll cover how to understand the benefits and cost of migrating to ARM in AWS EKS, including savings and practical guidelines for a transition.
Published Feb 9, 2025
Modern businesses are constantly looking for ways to optimize costs and improve cloud resource efficiency. AWS offers ARM-based servers (Graviton2 and Graviton3), which deliver significant advantages compared to traditional AMD or Intel processors. In this article, we’ll discuss how to migrate to ARM in your AWS EKS cluster, as well as analyze the benefits and cost savings of such a move.
Cost Savings: - Graviton-based instances provide up to 40% cost savings compared to equivalent AMD or Intel instances. - This reduction is achieved through lower hourly rates and higher performance efficiency.
Improved Performance: - ARM instances deliver up to 25% higher performance for many workloads, including microservices, CI/CD pipelines, and data processing.
Energy Efficiency: - ARM processors consume less power, making them more eco-friendly for large compute clusters.
Compatibility with Modern Workloads: - Graviton fully supports Kubernetes, enabling seamless integration into EKS clusters, especially with modern containerized applications.
| Parameter | m6g.2xlarge (ARM) | m6a.2xlarge (AMD) | Savings (%) | m6g.4xlarge (ARM) | m6a.4xlarge (AMD) | Savings (%) | m6g.8xlarge (ARM) | m6a.8xlarge (AMD) | Savings (%) |
|---|---|---|---|---|---|---|---|---|---|
| vCPUs | 8 | 8 | 16 | 16 | 32 | 32 | |||
| Memory (GB) | 32 | 32 | 64 | 64 | 128 | 128 | |||
| Performance (RPS) | 2,000 | 1,800 | 4,000 | 3,600 | 8,000 | 7,200 | |||
| Instances for 10,000 RPS | 5 | 6 | 3 | 3 | 2 | 2 | |||
| On-Demand Hourly Cost (eu-central-1) | $0.368 | $0.414 | 11.11% | $0.736 | $0.6984 | 11.11% | $1.472 | $1.656 | 11.11% |
| On-Demand Hourly Cost (eu-west-1) | $0.344 | $0.3852 | 12.70% | $0.6880 | $0.6566 | 12.70% | $1.376 | $1.541 | 12.70% |
| 1-Year Reserved (All Upfront) Hourly Cost (eu-central-1 | $0.185 | $0.237 | 21.94% | $0.370 | $0.474 | 21.94% | $0.740 | $0.948 | 21.94% |
| 1-Year Reserved (All Upfront) Hourly Cost (eu-west-1) | $0.173 | $0.222 | 22.07% | $0.346 | $0.444 | 22.07% | $0.692 | $0.888 | 22.07% |
| 3-Year Reserved (All Upfront) Hourly Cost (eu-central-1) | $0.145 | $0.190 | 23.68% | $0.290 | $0.380 | 23.68% | $0.580 | $0.760 | 23.68% |
| 3-Year Reserved (All Upfront) Hourly Cost (eu-west-1) | $0.176 | 23.30% | $0.270 | 23.30% | $0.540 | $0.704 | 23.30% | ||
| Spot Average Hourly Cost (eu-central-1) | $0.090 | $0.122 | $0.180 | $0.244 | 26.23% | $0.360 | $0.488 | 26.23% | |
| $0.084 | $0.114 | 26.32% | $0.168 | $0.228 | $0.456 | 26.32 |
Savings Percentage: Calculated as
(AMD Price - ARM Price) / AMD Price × 100 for each pricing model and instance type.Savings Range:
**On-Demand**: 11%–13%.
**Reserved (1-Year)**: ~22%.
**Reserved (3-Year)**: ~23%.
**Spot**: ~26%.
Spot Pricing: Offers the highest percentage savings, ideal for temporary or fault-tolerant workloads.
Performance (RPS): Estimated based on typical benchmarks. Actual performance may vary depending on workload characteristics.
Pricing: Accurate as of December 2024 for the eu-central-1 (Frankfurt) and eu-west-1 (Ireland) regions. For the latest rates, refer to the AWS Pricing Calculator.
Energy Efficiency: ARM-based instances (Graviton2) typically consume less power compared to AMD EPYC, which can significantly affect the total cost of ownership and environmental impact.
Compatibility: ARM instances may require recompilation or adaptation of applications and containers to support the
arm64 architecture.ARM (Graviton2): Ideal for workloads optimized for ARM or capable of being adapted, providing superior energy efficiency and cost savings.
AMD (EPYC): Better suited for workloads requiring immediate compatibility without modifications.
Assess Compatibility: - Ensure your applications and libraries are compatible with ARM architecture. Use container images tagged with
arm64.Prepare ARM Instances: - Add ARM-based instances (e.g.,
Graviton2 or Graviton3) to your EKS cluster. - Configure separate node groups or provisioning tools like Karpenter for ARM-based nodes.Test Workloads: - Deploy applications in a test environment on ARM instances to validate performance and stability.
Update CI/CD Pipelines: - Modify CI/CD pipelines to build ARM-compatible images. For example, in Docker, use the flag
--platform linux/arm64.Monitor and Optimize: - Use Amazon CloudWatch or Prometheus to track performance and ensure applications are functioning as expected.
Scenario:
A company migrated 100 microservices from AMD to ARM in its EKS cluster.
Results:
Cost Savings: Monthly costs reduced from $12,000 to $7,800, achieving 35% savings.
Performance Boost: CI/CD pipelines saw a 20% increase in efficiency.
Energy Reduction: Cluster energy consumption dropped by 30%, aligning with the company's sustainability goals.
ROI: Migration costs were recouped within 4 months due to reduced instance expenses.
If your applications support ARM architecture or can be easily adapted, transitioning to ARM in AWS EKS offers financial and environmental benefits. Even a partial migration to ARM instances can optimize costs and enhance performance.
Amazon Graviton3 processors offer significant improvements over their predecessors, providing enhanced performance and cost efficiency compared to AMD EPYC processors.
Performance Enhancements:
In a comprehensive benchmark involving 94 tests, Graviton3 instances secured the highest number of first-place finishes, outperforming both Intel Xeon and AMD EPYC instances.PhoronixGraviton3 demonstrated superior performance in high-performance computing (HPC) workloads, particularly in memory bandwidth-intensive applications, due to its advanced DDR5 memory support.PhoronixCost Efficiency:
AWS claims that Graviton processors deliver up to 40% better price-performance compared to comparable x86 instances.VantageIn specific workloads, such as Apache Kafka, Graviton3 instances have been identified as the price-performance leader, offering substantial cost savings over AMD EPYC instances.The New StackPricing Comparison:
Assuming that Graviton3 instances are priced similarly to Graviton2, we can estimate the following On-Demand hourly costs for the eu-west-1 (Ireland) region:
m6g.2xlarge (ARM/Graviton2): $0.3440
m6a.2xlarge (AMD EPYC): $0.3940
If Graviton3 offers a 10% performance improvement over Graviton2, the effective cost per unit of performance would be:
Graviton3 (estimated): $0.3440 / 1.10 ≈ $0.3127
This suggests that Graviton3 provides better performance per dollar compared to AMD EPYC instances.
Migrating from AMD to ARM in AWS EKS is a strategic move for companies aiming to improve their cloud infrastructure efficiency. With lower costs, higher performance, and support for modern technologies, ARM processors are a compelling choice for the future of cloud computing.
Transitioning to Graviton3 instances can lead to enhanced performance and cost savings, especially for workloads optimized for ARM architecture. However, it's crucial to assess application compatibility and potential migration efforts.