Amazon Web Services (AWS), a subsidiary of Amazon.com, has officially announced the general availability of its new Amazon Elastic Compute Cloud (EC2) R9g and R9gd instance families. Powered by the newly introduced AWS Graviton5 processors—touted by the company as the most energy-efficient silicon ever designed by AWS—these memory-optimized instances mark a significant generational leap in cloud computing infrastructure. The release responds to an escalating enterprise demand for higher performance efficiency, lower energy consumption, and mathematically proven security architectures in large-scale data processing.
The R9g instances are engineered specifically to cater to memory-intensive workloads, delivering up to a 25% improvement in compute performance over their predecessors, the Graviton4-powered R8g instances. According to AWS technical briefs, the latest generation provides higher performance per virtual CPU (vCPU), enhanced memory bandwidth, superior network and Amazon Elastic Block Store (EBS) throughput, and an expanded Level 3 (L3) cache. These architectural enhancements collectively allow organizations to handle heavier computational loads while simultaneously reducing their overall carbon footprint and energy expenditure.
Background Context and Evolution of AWS Silicon
The launch of the R9g and R9gd instances represents the latest milestone in Amazon’s sustained multi-year transition toward custom, in-house silicon. For decades, cloud providers relied predominantly on off-the-shelf central processing units manufactured by traditional semiconductor giants. However, rising power densities, thermal management challenges, and the unique performance bottlenecks of hyperscale cloud environments drove companies like Amazon to design application-specific integrated circuits tailored expressly for virtualized architectures.
The journey began in earnest with the acquisition of Annapurna Labs in 2015, which laid the groundwork for the Nitro system and the subsequent development of the custom ARM-based Graviton processor line. The first-generation Graviton processor, introduced in 2018, was primarily targeted at scale-out workloads with modest performance requirements. Subsequent iterations—Graviton2 in 2019, Graviton3 in 2021, and Graviton4 in 2024—each brought exponential leaps in performance, energy efficiency, and feature sets.
With Graviton5, AWS has focused heavily on refining the microarchitecture to optimize instruction execution, cache efficiency, and power gating. This steady evolution has transformed AWS from a consumer of merchant silicon into one of the world’s leading semiconductor designers, forcing traditional chipmakers to accelerate their own server-class roadmaps.
Chronology of the Graviton Ecosystem and Recent Milestones
The path to the general availability of R9g instances followed a carefully orchestrated sequence of hardware and software deployments across the AWS ecosystem:
- 2018: AWS introduces the first-generation Graviton processor, establishing the viability of ARM architecture in the cloud.
- 2019–2021: The release of Graviton2 and Graviton3 broadens the appeal of custom silicon, driving widespread adoption across relational databases, caching layers, and high-performance computing clusters.
- Early 2024: AWS rolls out C9g and M9g compute- and general-purpose instances powered by early iterations of next-generation architectures, introducing the pioneering Nitro Isolation Engine to enterprise customers.
- Late 2024: Graviton4-powered R8g instances establish a new benchmark for memory-optimized cloud computing.
- September 2025: AWS officially announces the general availability of R9g and R9gd instances powered by Graviton5 processors, alongside the commercial deployment of advanced Instance Bandwidth Configuration and expanded formal verification standards for cloud hypervisors.
Detailed Workload Suitability and Performance Profiles
The R9g instances are tailored for applications that demand high throughput and low latency from memory subsystems. Primary use cases include distributed relational databases, in-memory caching engines such as Valkey, Redis, and Memcached, as well as real-time big data analytics pipelines. Furthermore, the instances fully support Linux-based environments, containerized deployments running on Kubernetes, Docker, Amazon Elastic Kubernetes Service (EKS), and Amazon Elastic Container Service (ECS). Developers working with modern and legacy programming languages—including C/C++, Rust, Go, Java, Python, .NET Core, Node.js, Ruby, and PHP—can migrate to the new architecture seamlessly, often requiring little to no code modification.
For applications requiring high-speed temporary storage, the R9gd variants incorporate local NVMe-based solid-state drive (SSD) block-level storage. These instances are ideal for open-source databases, distributed analytics engines requiring rapid scratch space, large in-memory database transaction logs, and extensive caching layers that benefit from direct-attached storage with minimal input/output latency.
Hardware Innovations: Nitro System and the Nitro Isolation Engine
A foundational pillar of the R9g and R9gd architecture is their integration with the AWS Nitro System. By offloading virtualization, storage, and networking functions to dedicated hardware and firmware modules, the Nitro System frees up host CPU cores for customer workloads, delivering near-bare-metal performance while maintaining strict security boundaries.
Building upon the security framework introduced earlier in the year with the C9g and M9g instances, the R9g family features the Nitro Isolation Engine (NIE). The NIE is a purpose-built component responsible for enforcing absolute isolation between virtual machines. It achieves this by rigorously mediating all access to virtual machine memory, CPU register states, and I/O devices through a minimal, highly audited set of application programming interfaces (APIs).
Crucially, AWS has employed formal verification—a rigorous mathematical technique used to prove the correctness of hardware and software systems against formal specifications—to validate the NIE. This methodology goes far beyond conventional testing by mathematically demonstrating that the hypervisor behaves precisely as intended across all possible states, rather than just in pre-defined test scenarios. As a result, the AWS Nitro System stands as the industry’s first formally verified cloud hypervisor, establishing a new benchmark for provable cloud security and multi-tenant isolation.
Comprehensive Technical Specifications
To accommodate a wide array of enterprise architectures, the R9g and R9gd instances are available in 11 distinct sizing tiers, scaling from nimble single-core development environments to massive, bare-metal enterprise clusters.
R9g Instance Specifications (EBS-Only)
- r9g.medium: 1 vCPU, 8 GiB Memory, EBS-Only Storage, Up to 15 Gbps Network Bandwidth, Up to 12 Gbps EBS Bandwidth
- r9g.large: 2 vCPUs, 16 GiB Memory, EBS-Only Storage, Up to 15 Gbps Network Bandwidth, Up to 12 Gbps EBS Bandwidth
- r9g.xlarge: 4 vCPUs, 32 GiB Memory, EBS-Only Storage, Up to 15 Gbps Network Bandwidth, Up to 12 Gbps EBS Bandwidth
- r9g.2xlarge: 8 vCPUs, 64 GiB Memory, EBS-Only Storage, Up to 17 Gbps Network Bandwidth, Up to 12 Gbps EBS Bandwidth
- r9g.4xlarge: 16 vCPUs, 128 GiB Memory, EBS-Only Storage, Up to 17 Gbps Network Bandwidth, Up to 12 Gbps EBS Bandwidth
- r9g.8xlarge: 32 vCPUs, 256 GiB Memory, EBS-Only Storage, 17 Gbps Network Bandwidth, 12 Gbps EBS Bandwidth
- r9g.12xlarge: 48 vCPUs, 384 GiB Memory, EBS-Only Storage, 25 Gbps Network Bandwidth, 18 Gbps EBS Bandwidth
- r9g.16xlarge: 64 vCPUs, 512 GiB Memory, EBS-Only Storage, 34 Gbps Network Bandwidth, 24 Gbps EBS Bandwidth
- r9g.24xlarge: 96 vCPUs, 768 GiB Memory, EBS-Only Storage, 50 Gbps Network Bandwidth, 36 Gbps EBS Bandwidth
- r9g.48xlarge: 192 vCPUs, 1,536 GiB Memory, EBS-Only Storage, 100 Gbps Network Bandwidth, 72 Gbps EBS Bandwidth
- r9g.metal-48xl: 192 vCPUs, 1,536 GiB Memory, EBS-Only Storage, 100 Gbps Network Bandwidth, 72 Gbps EBS Bandwidth
R9gd Instance Specifications (Featuring Local NVMe SSD Storage)
- r9gd.medium: 1 vCPU, 8 GiB Memory, 1 x 59 GB NVMe SSD, Up to 15 Gbps Network Bandwidth, Up to 12 Gbps EBS Bandwidth
- r9gd.large: 2 vCPUs, 16 GiB Memory, 1 x 118 GB NVMe SSD, Up to 15 Gbps Network Bandwidth, Up to 12 Gbps EBS Bandwidth
- r9gd.xlarge: 4 vCPUs, 32 GiB Memory, 1 x 237 GB NVMe SSD, Up to 15 Gbps Network Bandwidth, Up to 12 Gbps EBS Bandwidth
- r9gd.2xlarge: 8 vCPUs, 64 GiB Memory, 1 x 474 GB NVMe SSD, Up to 17 Gbps Network Bandwidth, Up to 12 Gbps EBS Bandwidth
- r9gd.4xlarge: 16 vCPUs, 128 GiB Memory, 1 x 950 GB NVMe SSD, Up to 17 Gbps Network Bandwidth, Up to 12 Gbps EBS Bandwidth
- r9gd.8xlarge: 32 vCPUs, 256 GiB Memory, 1 x 1,900 GB NVMe SSD, 17 Gbps Network Bandwidth, 12 Gbps EBS Bandwidth
- r9gd.12xlarge: 48 vCPUs, 384 GiB Memory, 3 x 950 GB NVMe SSD, 25 Gbps Network Bandwidth, 18 Gbps EBS Bandwidth
- r9gd.16xlarge: 64 vCPUs, 512 GiB Memory, 1 x 3,800 GB NVMe SSD, 34 Gbps Network Bandwidth, 24 Gbps EBS Bandwidth
- r9gd.24xlarge: 96 vCPUs, 768 GiB Memory, 3 x 1,900 GB NVMe SSD, 50 Gbps Network Bandwidth, 36 Gbps EBS Bandwidth
- r9gd.48xlarge: 192 vCPUs, 1,536 GiB Memory, 3 x 3,800 GB NVMe SSD, 100 Gbps Network Bandwidth, 72 Gbps EBS Bandwidth
- r9gd.metal-48xl: 192 vCPUs, 1,536 GiB Memory, 3 x 3,800 GB NVMe SSD, 100 Gbps Network Bandwidth, 72 Gbps EBS Bandwidth
Additionally, both instance families introduce Instance Bandwidth Configuration (IBC), enabling administrators to dynamically adjust network and EBS bandwidth allocations by up to 25%. This fine-grained control allows database administrators and systems engineers to tune performance profiles to match the exact input/output characteristics of their workloads.
Migration Pathways, Operating Systems, and Developer Tools
AWS has engineered the R9g family to ensure frictionless migration for organizations currently operating on R8g or earlier ARM-based architectures. For the vast majority of applications, transitioning to R9g requires no source code modifications. Multi-architecture container images compiled for the ARM64 instruction set execute natively and efficiently across Amazon EKS, Amazon ECS, and standard Kubernetes distributions.
Supported operating systems at launch include Amazon Linux 2023, Amazon Linux 2, Ubuntu 22.04 and newer releases, Red Hat Enterprise Linux (RHEL) 8.4 and later, SUSE Linux Enterprise Server (SLES) 15 SP3 and above, as well as Debian 12 and other prominent Linux distributions.
To facilitate adoption, AWS provides a comprehensive suite of migration and optimization resources. The AWS Graviton Getting Started Guide outlines best practices for compiling and tuning software applications. The Graviton Savings Dashboard enables finance and operations teams to monitor cost optimizations in real time, while AWS Transform automates the conversion and modernization of legacy Java applications migrating from traditional x86 architectures to ARM-based processors.
Market Implications and Economic Analysis
Industry analysts view the rollout of Graviton5 and the R9g instance family as a strategic move that further strengthens Amazon’s competitive moat in cloud infrastructure. By controlling the silicon design process, AWS can decouple its pricing and performance trajectories from external semiconductor manufacturers, passing substantial cost savings along to enterprise customers.
As enterprises face mounting pressure to optimize IT budgets while scaling data processing capabilities, the combination of improved vCPU performance, reduced energy consumption, and flexible bandwidth allocation makes Graviton5 an increasingly compelling default choice for cloud-native software development. Furthermore, the introduction of mathematically proven hypervisor security via the Nitro Isolation Engine addresses enterprise concerns regarding multi-tenant isolation in highly regulated sectors such as finance, healthcare, and government services.
Pricing, Regional Availability, and Access
Amazon EC2 R9g and R9gd instances are available immediately across select AWS cloud regions, including US East (N. Virginia), US East (Ohio), US West (Oregon), and Europe (Frankfurt).
Customers can procure the new instances through multiple purchasing models, including On-Demand Instances, Savings Plans, Spot Instances, Dedicated Instances, and Dedicated Hosts. Comprehensive pricing structures and regional breakdowns are detailed on the official Amazon EC2 pricing portal. Engineering teams can begin provisioning R9g and R9gd workloads directly through the Amazon EC2 management console or via infrastructure-as-code tooling supported by the broader AWS ecosystem.
