When Amazon Web Services (AWS) launched its new Arm-based processors, some circles believed it was a game-changer for the public cloud markets. To begin with, it was the first time Arm architecture would roll out for enterprise-grade utility, and at a colossal scale.
Arm processors had only run on smaller, less demanding devices such as iPhones. So why adopt it for much more demanding workloads in cloud services?
This post looks at what AWS Graviton is, if Graviton2 processors are better than the first-generation Graviton A1 cores, and what it all means for you.
What Is AWS Graviton?
AWS Graviton is a series of server processors that AWS released in 2018 based on Arm architecture for customers of its Amazon Elastic Compute Cloud (EC2) virtual machine instances. The first generation AWS Graviton processors featured custom silicon and 64-bit Neoverse cores.
The EC2 A1 instances support Arm-based applications such as web servers, caching fleets, distributed data centers, and containerized microservices.
They settled on using an open Arm architecture. That meant saving costs by not creating a new chip from scratch. Instead, they took an existing Arm architecture and customized it for how AWS EC2 instances work.
Why did AWS create Graviton processors?
Talking to NewStack towards the end of 2020, David Brown, Vice President of EC2 at AWS, revealed an interesting perspective.
He said they noted a staggering number of Amazon EC2 customers, both large and small, were barely using their EC2 capacities. So after listening to customers such as SmugMug and Flickr, AWS switched from the X86-64 family of processors for their servers for several reasons.
AWS wanted to:
- Offer more choice in terms of selection of EC2 instances for customers
- Target Arm-based applications such as web servers
- Provide high availability and security while reducing virtualization costs
- Align decent server performance with lower prices for customers
AWS likely also wanted an in-house family of server processors built to work how AWS works rather than depending on Intel and AMD for innovation.
Are AWS Graviton processors any good?
Some circles felt the first-generation AWS Graviton processors played second fiddle to AMD and Intel processors at the time. But over time, the processors proved slightly better than X86-based processors for servers.
Ultimately, Arm processors have lower power consumption compared to X86 cores, for example. That may be one proposition that AWS had been going for, so it could trickle savings down to EC2 pricing.
But enterprise customers wanted in on the Graviton architecture, too. So in May 2020, AWS announced AWS Graviton2 with the promise to handle much more demanding workloads than before.
Research Report
FinOps In The AI Era: A Critical Recalibration
What 475 executives told us about AI and cloud efficiency.
AWS Graviton Vs. Graviton2: What Are The Differences?
At launch, AWS promised Graviton2 offered 40% “better price-performance than comparable” X86 processors and 7X better than first-generation AWS Graviton processors. The new-generation processors should also offer 4X compute cores, a memory that’s 5X faster, and 2X as large caches as Graviton1 processors.
With Graviton2, AWS also made some key improvements to empower developers to create cloud-native apps that can run securely and at scale. That includes the always-on 256-bit DRAM encryption.
Here is a closer look at the differences between first-generation AWS Graviton and Graviton2 processors.
1. Compute cores and architecture
The original Graviton processor uses 16 Cortex-A72 cores based on a 16nm process. Graviton2 jumps to 64 Neoverse N1 cores built on a 7nm process, delivering significantly more parallel processing power. That makes Graviton2 suitable for compute-heavy workloads like batch processing, video encoding, and large-scale application serving that the first-generation chip struggled with.
2. Memory and cache
Graviton2 introduced support for DDR4-3200 memory across eight channels, a meaningful upgrade over the original Graviton’s more limited memory subsystem. Each core also accesses a larger private L2 cache and a shared last-level cache, which reduces latency for memory-intensive operations like in-memory databases and real-time analytics.
3. Networking and I/O
First-generation Graviton instances topped out at 10 Gbps of network bandwidth. Graviton2-based instances like the C6gn can push up to 100 Gbps with Elastic Fabric Adapter (EFA) support, opening the door to high-performance computing and tightly coupled distributed systems.
4. Security
Graviton2 adds always-on 256-bit DRAM encryption, pointer authentication for code-reuse attack mitigation, and support for encrypted EBS volumes with minimal performance overhead. The original Graviton had none of these hardware-level security features.
5. Service breadth
Graviton2 is not limited to just powering EC2 instances. It runs across Amazon ElastiCache, Amazon RDS, Amazon EKS, Amazon EMR, and AWS Lambda, among others. First-generation Graviton was confined to EC2 A1 instances only.
Yet, that’s not all. AWS has since released Graviton3 (in 2022), Graviton4 (in November 2023), and Graviton5 (in December 2025) processors.
Graviton | Graviton2 | Graviton3 | Graviton4 | Graviton5 | |
Launch Year | 2018 | 2019 | 2022 | 2023 | 2025 |
Core Type | Cortex A72 | Neoverse N1 | Neoverse V1 | Neoverse V2 | Neoverse V3 |
Number of Cores | 16 | 64 | 64 | 96 | 192 |
Clock Rate | 2.3 GHz | 2.5 GHz | 2.6 GHz | ~2.8 GHz | TBA |
Technology Node | 16 nm | 7 nm | 5 nm | 4 nm | 3 nm |
Memory Channels | — | 8× DDR4 | 8× DDR5-4800 | 12× DDR5-5600 | DDR5 (enhanced) |
Key Features | Basic ARMv8-A support, Neon, CRC | ARMv8.2-A, Neon, Dotprod, FP16, DRAM encryption | ARMv8.4-A, SVE, BFloat16, Pointer Authentication | ARMv9.0-A, BTI, Confidential Computing | ARMv9, 5× larger L3 cache, Nitro Isolation Engine |
Performance | Suitable for basic and non-performance-critical workloads | Up to 7× better than Graviton1 and ~40% better price/performance vs x86 | 25% better compute, 2× crypto, 3× ML vs Graviton2 | Up to 40% faster databases, 30% faster web apps, 45% faster Java apps | 25% better compute, 30% faster databases, 35% faster web apps vs Graviton4 |
The table highlights how AWS Graviton processors have improved over time in terms of performance, efficiency, and support for advanced computing tasks.
When Should You Use AWS Graviton Processors?
Use AWS Graviton and Graviton2 for web servers, log processing, video encoding, electronic design automation, and machine learning based on a CPU interface.
Keep in mind that if you currently use X86-based servers, you would need to re-architect your application to run on the Arm architecture. For the trouble, you would see a significant reduction in price performance, but over time.
On that note, to really answer this question, you’ll want to first know exactly where AWS Graviton processors work.
Which Amazon EC2 instances are powered by Graviton processors?
Below you’ll find an overview of Amazon EC2 instances using various generations of AWS Graviton processors, along with a description and their best use case:
Powered by AWS Graviton2
- T4g – This instance type supplies excellent price performance for burstable general-purpose workloads. Ideal use cases include running large-scale microservices and small to medium databases.
- C6g, C6gd, C6gn – AWS designed these instances to handle compute- and network-intensive workloads. They are suitable for applications such as high-performance computing and video encoding.
- X2gd – This option provides the lowest cost per GiB of memory. It is ideal for memory-intensive workloads like open-source databases and real-time analytics.
- Im4gn – This instance type is optimized for storage-intensive workloads, and is perfect for SQL and NoSQL databases, search engines, and analytics.
- Is4gen – This option offers the lowest cost per TB of SSD storage, suitable for databases, search engines, and large file systems.
- G5g – If you are looking for an ideal option for graphics applications, including Android game streaming and machine learning inference, the G5g option may help.
Powered by AWS Graviton3
- M7g, M7gd – These instances are ideal for general-purpose workloads that require balanced compute, memory, and networking power. Ideal use cases for them include running application servers, midsize data stores, and microservices.
- C7g, C7gd, C7gn – These are best used for compute-intensive workloads such as high-performance computing, video encoding, gaming, and CPU-based machine learning inference acceleration.
- R7g, R7gd – These two are ideal for memory-intensive workloads, including open-source databases (MySQL, MariaDB, PostgreSQL) and in-memory caches (Redis, KeyDB, Memcached).
Powered by AWS Graviton4
- R8g – Tailored for memory-intensive workloads, such as high-performance databases and real-time big data analytics.
- M8g – General-purpose instances delivering up to 30% better price-performance over M7g for application servers, mid-size databases, and enterprise workloads.
- C8g – Compute-optimized instances suited for batch processing, scientific modeling, and machine learning inference.
Powered by AWS Graviton5 (Preview)
- M9g – The newest general-purpose instances (available in preview as of early 2026), offering up to 25% better compute performance than Graviton4-based M8g instances. AWS has also announced plans for C9g (compute-optimized) and R9g (memory-optimized) instances in 2026.
Which AWS services support Graviton processors today?
Here’s a quick rundown of AWS services that support Graviton processors and what you can do with them:
- Amazon EC2 – This is like the backbone of cloud computing, giving you virtual servers to run your applications smoothly.
- Amazon ElastiCache – Think of it as a turbocharger for your database, speeding up data retrieval from your databases. Think of it as a turbocharger for your database, speeding up data retrieval from your databases.
- Amazon Elastic Kubernetes Service (EKS) – This is like a conductor for your container orchestras, helping manage and orchestrate your containerized applications.
- Amazon Aurora – A super-efficient, high-performance, and scalable relational database service that makes handling your data a breeze. It is compatible with MySQL and PostgreSQL.
- Amazon Relational Database Service (RDS) – This service simplifies setting up, operating, and scaling a relational database in the cloud, and taking care of tedious tasks like backups and scaling.
- Amazon MemoryDB for Redis – Perfect for storing and retrieving data super quickly, it’s like a high-speed memory lane for your apps. It delivers a Redis-compatible, in-memory database service for applications requiring microsecond latency.
- Amazon OpenSearch – This AWS service provides a suite of search and analytics services for analyzing large volumes of data quickly and at scale. Think of it as your cloud library’s search engine, making data search and analysis simple and fast.
- Amazon EMR – This service is designed to handle big data, helping you process vast amounts of data easily. It uses open-source tools such as Apache Spark and Hadoop.
- AWS Lambda – This serverless service runs code in response to events and automatically manages computing resources required by that code.
- AWS Fargate – Fargate is a serverless compute engine for containers that works with both Amazon Elastic Container Service (ECS) and Amazon Elastic Kubernetes Service (EKS). Fargate takes away the hassle of managing servers to run containers.
With Graviton processors, these services make running certain tasks in the cloud more efficient and cost-effective. So, whether you’re managing data, running applications, or dealing with big data, there’s a Graviton-powered service to make your life easier.
What Are The Benefits Of Using AWS Graviton?
The most significant AWS Graviton benefits are reduced costs, low latency, better scalability, improved availability, and increased security.
1. Cost-effective
Graviton-based instances cost up to 20% less per hour than comparable x86 instances. When you factor in the performance gains, AWS estimates up to 40% better price-performance overall. For compute-heavy workloads, organizations have reported total infrastructure savings between 20% and 50% after migrating.
2. Eco-system support
AWS Graviton and Graviton2 are based on the 64-bit Arm Neoverse core architecture. Several Linux-based operating systems support the configuration. They include Amazon Linux 2, SUSE, and Red Hat. That provides more choice to customers.
3. Effective CPU power
Each Graviton vCPU maps to a dedicated physical core rather than a shared hyperthread. That means more predictable performance under load, which is particularly relevant for latency-sensitive applications like real-time APIs or database queries.
4. Built for general purpose
AWS Graviton cores are also built to improve efficiency in servers, mid-size data-storing processes, micro-services, and cluster computing.
5. Offers burstable workload
They provide users with an extensive set of burstable workload services such as scale microservices, small and medium database services, virtual desktops, and a selection of applications suitable for critical business.
6. Build on a computer-intensive model
The processors are also built on a computer-intensive model like HD video performance computing, encoding videos, gaming, and CPU-based computer learning processes.
7. Offers enhanced networking
Expect support for a C6gn network at 100 Gbps networking capabilities of the Elastic Fabric Operator (EFO).
How Does AWS Graviton Billing Work?
After learning how AWS Graviton works and its computing benefits, here’s a look at Graviton and Graviton2 pricing:
- General Purpose
- Current generation computer-optimized
- Graphic processing unit instances current generation pricing
- Graphic processing unit instances previous generation pricing
Each part describes hourly price ranges.
For example: If you use on-demand m6g.xlarge for an hour, expect to see a $0.154 bill. If you opt for the EMR service, you are looking at a $0.039 bill.
Software and pricing information last verified April 2026. Features, pricing, and availability may have changed. Please verify current details with AWS before making decisions.
Price model for general purpose
On-Demand Price | EMR Price | |
m6g.xlarge | $0.154 per hour | $0.039 per hour |
m6g.2xlarge | $0.308 per hour | $0.154 per hour |
m6g.4xlarge | $0.616 per hour | $0.154 per hour |
m6g.8xlarge | $1.232 per hour | $0.308 per hour |
m6g.12xlarge | $1.848 per hour | $0.462 per hour |
m6g.16xlarge | $2.464 per hour | $0.616 per hour |
m6gd.xlarge | $0.1808 per hour | $0.0452 per hour |
m6gd.2xlarge | $0.3616 per hour | $0.0904 per hour |
m6gd.4xlarge | $0.7232 per hour | $0.1808 per hour |
m6gd.8xlarge | $1.4464 per hour | $0.3616 per hour |
m6gd.12xlarge | $2.1696 per hour | $0.5424 per hour |
m6gd.16xlarge | $2.8928 per hour | $0.7232 per hour |
m5.xlarge | $0.192 per hour | $0.048 per hour |
m5.2xlarge | $0.384 per hour | $0.096 per hour |
m5.4xlarge | $0.768 per hour | $0.192 per hour |
m5.8xlarge | $1.536 per hour | $0.27 per hour |
m5.12xlarge | $2.304 per hour | $0.27 per hour |
m5.16xlarge | $3.072 per hour | $0.27 per hour |
m5.24xlarge | $4.608 per hour | $0.27 per hour |
Current generation computer-optimized pricing
c6g.xlarge | $0.136 per hour | $0.034 per hour |
c6g.2xlarge | $0.272 per hour | $0.068 per hour |
c6g.4xlarge | $0.544 per hour | $0.136 per hour |
c6g.8xlarge | $1.088 per hour | $0.272 per hour |
c6g.12xlarge | $1.632 per hour | $0.408 per hour |
c6g.16xlarge | $2.176 per hour | $0.544 per hour |
c6gd.xlarge | $0.1536 per hour | $0.0384 per hour |
c6gd.2xlarge | $0.3072 per hour | $0.0768 per hour |
c6gd.4xlarge | $0.6144 per hour | $0.1536 per hour |
c6gd.8xlarge | $1.2288 per hour | $0.3072 per hour |
c6gd.12xlarge | $1.8432 per hour | $0.4608 per hour |
c6gd.16xlarge | $2.4576 per hour | $0.6144 per hour |
c6gn.xlarge | $0.1728 per hour | $0.0432 per hour |
c6gn.2xlarge | $0.3456 per hour | $0.0864 per hour |
c6gn.4xlarge | $0.6912 per hour | $0.1728 per hour |
c6gn.8xlarge | $1.3824 per hour | $0.3456 per hour |
c6gn.12xlarge | $2.076 per hour | $0.5184 per hour |
c6gn.16xlarge | $2.7646 per hour | $0.6912 per hour |
c5.xlarge | $0.17 per hour | $0.04 per hour |
c5.2xlarge | $0.34 per hour | $0.085 per hour |
Graphics processing unit (GUP) instances current generation pricing
p3.2xlarge | $3.06 per hour | $0.27 per hour |
p3.8xlarge | $12.24 per hour | $0.27 per hour |
p3.16xlarge | $24.48 per hour | $0.27 per hour |
g4dn.xlarge | $0.526 per hour | $0.132 per hour |
g4dn.2xlarge | $0.752 per hour | $0.188 per hour |
g4dn.4xlarge | $1.204 per hour | $0.27 per hour |
g4dn.8xlarge | $2.176 per hour | $0.27 per hour |
g4dn.12xlarge | $3,912 per hour | $0.27 per hour |
g4dn.16xlarge | $4.352 per hour | $0.27 per hour |
g3.4xlarge | $1.14 per hour | $0.27 per hour |
g3s.xlarge | $0.75 per hour | $0.188 per hour |
Graphics processing unit (GUP) instances previous-generation pricing
p2.xlarge | $0.90 per hour | $0.225 per hour |
p2.8xlarge | $7.20 per hour | $0.27 per hour |
p2.16xlarge | $14.40 per hour | $0.27 per hour |
Graviton Benchmarks
Graviton4 vs. Intel and AMD
Independent benchmarks from Phoronix and OpenBenchmarking.org show Graviton4 (R8g/M8g instances) competing directly with AMD EPYC Genoa and Intel Xeon Sapphire Rapids in real-world workloads. In compilation tasks, Graviton4 delivers the best cost-per-run: approximately $0.186 for a Gem5 build, compared to $0.288 for AMD EPYC and $0.194 for Intel Xeon on equivalent instances. For 7-Zip compression, Graviton4 outperforms both x86 families. In OLTP database workloads, Graviton4 shows roughly 38% lower latency than Xeon and 20% lower than AMD EPYC.
The one area where x86 still holds an edge is certain single-threaded PostgreSQL workloads, where Intel and AMD retain a small advantage. For most cloud-native, multi-threaded applications, however, the Graviton price-performance gap continues to widen in AWS’s favor.
Graviton5 early benchmarks
Graviton5 (M9g instances, preview) pushes further with 192 cores and a 5x larger L3 cache than Graviton4. AWS reports up to 33% lower inter-core latency thanks to a redesigned internal layout, which should benefit latency-sensitive workloads like real-time bidding, financial modeling, and microservices architectures.
Choosing The Right Solution For Your Organization
AWS’s Arm-based Graviton processors offer cost savings, security, scalability, flexibility, and increased performance over X86 and X64-based processors — with each new generation widening the gap.
Graviton4 and Graviton5 EC2 instances support a broad array of capabilities, including native integration with services like RDS, EKS, Lambda, and Fargate. But because Graviton runs on Arm architecture, migrating from x86 requires testing your application stack for compatibility. Most modern Linux-based, containerized, and cloud-native workloads make the transition smoothly.
Want to switch over to AWS Graviton-powered services?
Here’s the thing.
Understanding your cloud spend (and seeing exactly how specific AWS services drive your product and feature costs) is the first step to deciding the best way to architect your infrastructure.
When you don’t have true visibility into your costs, you only know how much you pay every month, but not the specifics of who, what, and why your cloud bill is changing. That’s what conventional cloud cost management tools do.
Not CloudZero.
With CloudZero’s Cloud Cost Intelligence platform, you can:
- View per unit costs, such as Cost per Customer, Cost per Team, and Cost per Feature, to help you tell exactly why, who, and what is driving your cloud bill (and do something about it!)
- Allocate 100% of your cloud spend in minutes to hours no matter how complex your environment is so you can keep track of your full costs sooner rather than later
- Get a complete view of your cloud costs in a single source of truth, whether you use AWS, Azure, GCP, or Oracle Cloud — or even platforms such as Snowflake, MongoDB, Databricks, Datadog, New Relic, and more
- Get timely, noise-free, and context-rich cost anomaly alerts to help you fix issues that could lead to cost overruns
- CloudZero offers more, including intelligent budget tracking, accurate forecasting, and industry-leading Kubernetes cost analysis
But, don’t just take our word for it. Our customers, such as Drift and Upstart have saved over $2.4 million and $20 million, respectively. Here’s a chance as well. 
to experience CloudZero in action yourself.
FinOps In The AI Era: A Critical Recalibration
What 475 executives told us about AI and cloud efficiency.
