AWS launches Graviton5-powered R9g and R9gd EC2 instances with 25% better compute performance than R8g

AWS has released R9g and R9gd EC2 instances, powered by Graviton5 processors, as a direct upgrade to the R8g series. The new instances target memory-intensive workloads such as databases, in-memory caches, and real-time analytics, offering a 25% improvement in compute performance per vCPU. This performance boost is accompanied by hardware enhancements, including DDR5 memory running at 8800 MT/s, currently the fastest available in the cloud, and a significantly larger L3 cache, which improves data locality and reduces latency for memory-bound applications.

The R9g instances are designed for workloads that rely on EBS storage, while R9gd instances add local NVMe-based SSD storage, making them suitable for applications requiring fast, low-latency scratch space. Both instance types support Instance Bandwidth Configuration (IBC), a feature that allows engineers to adjust the allocation of bandwidth between EBS and VPC networking in 25% increments. This flexibility is particularly useful for workloads with specific bandwidth requirements, such as high-throughput databases or caching layers, where optimizing network and storage performance can directly impact application responsiveness.

Under the hood, Graviton5 processors introduce several architectural improvements over Graviton4, including up to 2x higher network and EBS bandwidth for the largest instance sizes and up to 3x higher packet-processing performance. These enhancements are delivered while maintaining energy efficiency, a key consideration for large-scale deployments. The instances also leverage the AWS Nitro System, with the Nitro Isolation Engine providing formally verified security isolation between virtual machines. This ensures that performance gains do not come at the cost of security or isolation, a critical factor for multi-tenant cloud environments.

For engineers already running workloads on R8g instances, migrating to R9g or R9gd should be straightforward, as the new instances maintain compatibility with existing Graviton-based applications. The performance improvements are likely to be most noticeable in latency-sensitive workloads, such as in-memory databases or real-time analytics, where faster memory and larger caches can reduce bottlenecks. However, the largest performance gains will be realized on the higher-end instance sizes, where network and EBS bandwidth are significantly increased. Teams should evaluate their workloads to determine whether the performance uplift justifies the cost, particularly for smaller instance sizes where the improvements may be less pronounced.