AWS Makes Graviton5-Powered EC2 R9g Instances Available for Memory-Heavy Workloads

Amazon Web Services has announced the general availability of Amazon EC2 R9g and R9gd instance types, powered by custom AWS Graviton5 processors. The new memory-optimized instances deliver up to 25% higher compute performance per vCPU compared to previous-generation Graviton4 R8g instances. Hardware upgrades include DDR5 memory rated at 8800 MT/s, a 5x larger L3 cache, up to 100 Gbps network bandwidth, and up to 72 Gbps dedicated Amazon EBS bandwidth on the largest sizes. The R9gd variant incorporates local NVMe-based solid-state storage. Both instance types also introduce Instance Bandwidth Configuration (IBC), which lets operators dynamically shift up to 25% of bandwidth allocation between VPC networking and EBS storage. For DevOps, SRE, and platform teams operating stateful infrastructure, memory speed and cache locality often dictate real-world tail latency more than raw core counts. High-traffic caching layers (such as Redis and Valkey), distributed big data analytics, and search clusters frequently hit memory bus saturation and I/O bottlenecks before exhausting CPU capacity. The jump to 8800 MT/s DDR5 memory combined with expanded L3 caching directly addresses these memory-access bottlenecks. Organizations can run larger working sets with tighter latency bounds, reducing the total node count needed to service peak request volumes. This release advances AWS's ongoing custom silicon strategy, completing the rollout of Graviton5 across core general compute (M9g), compute-optimized (C9g), and memory-optimized (R9g) families. Custom Arm-based processors have transitioned from budget alternatives to performance leaders in hyperscale cloud environments. As AI pipelines and real-time streaming engines demand higher data throughput between memory and processor cores, cloud providers are aggressively designing specialized silicon and memory architectures to outpace commodity hardware limits while meeting stricter datacenter power and efficiency mandates. In practice, teams already running Arm64-compatible Linux containers, Java runtimes, or database engines can execute near-drop-in instance migrations to capture immediate efficiency wins. Workloads that write heavily to ephemeral scratch disks—such as local query buffers or transient indexes—should target R9gd instances to leverage local NVMe throughput. Additionally, teams should benchmark the Instance Bandwidth Configuration feature to fine-tune network versus EBS allocations depending on whether nodes are network-bound (such as caching clusters) or storage-bound (such as write-heavy database replicas).