AWS Officially Launches EC2 R9g and R9gd Instances Based on Graviton5

Amazon EC2 has officially launched the memory-optimized instances R9g and R9gd based on the AWS Graviton5 processor. R9g offers up to 25% improved computing performance per vCPU compared to the previous generation R8g based on Graviton4, and features the most energy-efficient processor in AWS history.

Performance and Architecture Improvements

Graviton5 offers the following hardware improvements over Graviton4.

• Up to 25% higher computing performance per vCPU
• The fastest DDR5 8800 MT/s memory in the cloud (improved from the existing 5600 MT/s)
• 5x larger L3 cache for improved Data Locality
• Up to 2x increase in network and EBS bandwidth based on maximum instance size (100 Gbps network, 72 Gbps EBS for 48xlarge)
• Up to 3x improved packet processing performance

R9g and R9gd support Instance Bandwidth Configuration (IBC), allowing bandwidth allocation between Amazon EBS and VPC networking to be adjusted in 25% increments. This enables performance optimized for specific bandwidth requirements of database or caching workloads.

Enhanced Security and Isolation

All R9g and R9gd instances run on the AWS Nitro System, which offloads virtualization, storage, and networking to dedicated hardware. In particular, the Nitro Isolation Engine (NIE), first introduced in C9g and M9g instances, is applied, mathematically proving isolation between virtual machines through Formal Verification. NIE mediates VM memory, CPU register state, and I/O device access with a minimal API, thereby presenting a new standard for mathematically proven cloud security.

Instance Types and Use Cases

R9g instances are suitable for memory-intensive workloads such as databases, in-memory caches (Valkey, Redis, etc.), real-time big data analytics, and container-based applications (Kubernetes, Docker, etc.). R9gd instances provide the same computing and networking performance as R9g, while additionally including local NVMe SSD block storage. This is useful for distributed big data analytics, large-scale in-memory databases, and other scenarios requiring fast-latency temporary storage or caching. Both instances are offered in 11 sizes, ranging from medium to metal-48xl.