AWS Cost Efficiency
20
mins read
Choosing the Right AWS Graviton Instance for Your Workloads: A Cost Analysis
AWS Graviton Series Part 1
AWS Graviton processors optimize price performance for cloud workloads on Amazon EC2. Graviton2-based instances offer up to 40% better price performance than current x86 instances, benefiting workloads like application servers, microservices, video encoding, HPC, EDA, gaming, databases, in-memory caches, and CPU-based machine learning.[1]
In this blog post, we will explore the scenarios and use cases for Graviton Instances, as well as conduct a cost comparison between x86 and Graviton Instances. We will also discuss cost considerations for AWS Graviton Instances. This will help you make informed decisions when choosing the right instance type for your business workloads.
Given below are the use-cases/scenarios where one can use Graviton instances to achieve cost savings while meeting desired workload requirements.[3]
1. Scenario: Web Application Hosting
"WebTech Solutions Inc.," a web development company, currently utilizes EC2 On-Demand instances for hosting its portfolio of web applications and websites.
Their platform serves a diverse range of clients and requires reliable and scalable infrastructure to handle fluctuating traffic patterns.
WebTech Solutions employs 30 instances of t3.large, each with 2 vCPUs, 8 GiB of memory, and running on Amazon Linux 2.
As WebTech Solutions expands its client base and enhances its web applications, optimizing their cloud infrastructure becomes crucial for both performance and cost efficiency.
To address these requirements, WebTech can migrate to AWS General Purpose instances, specifically the M6g family.
Powered by AWS custom Graviton2 processors, the M6g instances offer a balance of compute, memory, and networking resources, making them well-suited for hosting web applications with varying workload demands.
2. Current Costs
3. Optimized Costs - WebTech Solutions plans to switch to M6g instances
4. Savings
1. Scenario: Batch Processing and Analytics
A data analytics firm, "DataMetrics," currently hosts its data processing platform on EC2 On-Demand instances.
The platform serves various clients and handles large-scale data analysis tasks, including batch processing and real-time analytics.
DataMetrics relies on 15 instances of c5.xlarge, each equipped with 4 vCPUs and 8 GiB of memory, running on Ubuntu 20.04 LTS. As DataMetrics scales its operations and processes increasingly large datasets, optimizing cloud infrastructure becomes critical for cost efficiency and performance.
To address these needs, DataMetrics should migrate to AWS Graviton Compute-Optimized instances, specifically the C6g family.
Powered by AWS Graviton2 processors, these instances offer superior performance for compute-intensive workloads at a lower cost. By transitioning to the C6g instances, DataMetrics expects to achieve significant cost savings while improving the performance of its data processing platform.
2. Current Costs
3. Optimized Costs - Data Metrics plans to switch to C6g (Graviton) instances
4. Savings
By migrating to AWS Graviton Compute-Optimized instances, DataMetrics can save approximately $3,672 annually, representing a reduction of around 20-25% in their infrastructure expenses.
Observing the performance of the new Graviton3-based C7g instances, they noted a 20-80% increase in performance and up to 35% reduced tail latencies.- Nick Tornow, Head of Platform, Twitter[3]
1. Scenario: AI Model Training
"AI Solutions Inc", a cutting-edge artificial intelligence startup, currently relies on EC2 On-Demand instances for training large-scale machine learning models.
Their AI models require extensive memory capacity to store and process massive datasets efficiently. AI Solutions utilizes 30 instances of r5.8xlarge, each with 32 vCPUs and 256 GiB of memory, running on Ubuntu 20.04 LTS.
As AI Solutions ventures into more ambitious AI projects and deals with increasingly complex datasets, optimizing their cloud infrastructure becomes essential for both performance and cost efficiency.
To address these requirements, AI Solutions can migrate to AWS Memory-Optimized instances, specifically the R6gd family.
Powered by AWS custom Graviton2 processors and featuring local NVMe storage, the R6gd instances offer exceptional memory performance and storage capabilities, ideal for memory-intensive AI workloads such as model training and inference.
2. Current Costs
3. Optimized Costs - AI Solutions plans to switch to R6gd instances
4. Savings
By migrating to AWS Memory-Optimized instances with local NVMe storage (R6gd), AI Solutions can save approximately $25,920 annually, enabling them to train more advanced AI models and handle larger datasets while optimizing their infrastructure costs.
We were able to transition our caching servers to r6gd.8xlarge instances with a simple recompile and are enjoying a 33% price benefit and improved latency. -Andrew Shieh, Director of Operations at SmugMug[3]
1. Scenario: Content Distribution and Media Streaming
"StreamNet Inc.," a leading provider of streaming media services, currently relies on EC2 On-Demand instances to distribute and stream media content to their global audience.
Their platform requires high storage capacity and throughput to deliver high-definition video content seamlessly. StreamNet utilizes 20 instances of i3.16xlarge, each with 64 vCPUs, 488 GiB of memory, and 8 x 7.6 TB NVMe SSD storage, running on Amazon Linux 2.
As StreamNet expands its content library and audience reach, optimizing their cloud infrastructure becomes paramount for both storage capacity and performance.
To meet these requirements, StreamNet can migrate to AWS Storage-Optimized instances, specifically the Im4gn family. Powered by AWS Graviton2 processors, the Im4gn instances offer superior storage performance and efficiency, ideal for storage-intensive workloads such as content distribution and media streaming.
2. Current Costs:
3. Optimized Costs - StreamNet plans to switch to Im4gn instances
4. Savings
By migrating to AWS Storage-Optimized instances powered by Graviton2 processors (Im4gn), StreamNet can save approximately $34,560 annually, ensuring they have the necessary storage capacity and performance to deliver high-quality streaming media content to their global audience while optimizing costs.
At Elastic, we tested the I4g instances, powered by AWS Graviton2 processors and we saw up to 35% better price-performance.-Yuvraj Gupta, Principal Product Manager, Elastic[3]
1. Scenario: AI/ML Model Training
"AI Inference Solutions Ltd.," a machine learning inference service provider, currently utilizes EC2 On-Demand instances for running inference tasks on trained machine learning models.
Their platform serves real-time predictions to various applications and requires high-speed inferencing capabilities. AI Inference Solutions employs 15 instances of p3.2xlarge, each equipped with NVIDIA Tesla V100 GPUs and 8 vCPUs, running on Ubuntu 20.04 LTS.
As AI Inference Solutions scales its inference services and handles more concurrent requests, optimizing their cloud infrastructure becomes crucial for both performance and cost efficiency.
To address these requirements, AI Inference Solutions can migrate to AWS Accelerated Computing instances, specifically the G5g family. Powered by AWS custom Graviton2 processors and featuring NVIDIA A100 Tensor Core GPUs, the G5g instances offer exceptional performance for machine learning inference tasks, enabling AI Inference Solutions to deliver real-time predictions with low latency and high throughput.
2. Current Costs
3. Optimized Costs - AI Inference Solutions plans to switch to G5g instances
4. Savings
By migrating to AWS Accelerated Computing instances with NVIDIA A100 GPUs (G5g), AI Inference Solutions can save approximately $17,280 annually.
Mircom modernized OpenGN’s single pane of glass and reduced infrastructure costs 30–40% using Amazon EC2 G5g Instances.-Tony Falbo, founder and CEO of Mircom.[4]
Consider checking AWS Graviton Instance page for Pricing and cost considerations.
When deciding between AWS Graviton instances (based on ARM architecture) and traditional x86-based instances (Intel/AMD), it's essential to consider various aspects such as pricing, performance, and workload compatibility. Below is a detailed comparison highlighting the key differences and advantages of each option.[5]
In summary, choosing the best AWS Graviton option needs a strategic approach. You have to consider how much it costs and how well it works for what you need. Graviton3 is great for handling lots of data quickly, especially for tasks like real-time analytics and managing databases. On the other hand, Graviton2 is good for different types of work and is both cost-effective and performs well.
It's important to understand how Graviton works with AWS services so you can make the most of it and improve your overall infrastructure. By utilizing AWS Graviton intelligently, businesses can not only reduce expenses but also enhance efficiency across a spectrum of digital tasks, contributing to overall operational excellence.
References
1. Enable Up to 40% Better Price Performance for Your Workloads with AWS Graviton2 Based Amazon EC2 Instances
2. aws-graviton-getting-started/managed_services.md at main
3. AWS Graviton - ARM Processor
4. Creating an Optimized Solution for Smart Buildings Using Amazon EC2 G5g Instances with Mircom’s OpenGN | Case Study | AWS
.png)
.png)
