Amazon has announced the general availability of the R9g and R9gd instance families, the first EC2 offerings powered by the Graviton5 processor. The new instances add up to 25% more compute per vCPU, DDR5‑8800 MT/s memory, a five‑fold larger L3 cache, and up to double the network and EBS bandwidth compared with the previous Graviton4‑based R8g line, while retaining the Nitro System’s isolation guarantees.
Key hardware improvements
- Compute uplift: Graviton5 delivers up to 25% higher per‑vCPU performance.
- Memory speed: DDR5 operates at 8800 MT/s, a significant jump from the 5600 MT/s of Graviton4.
- Cache size: L3 cache is five times larger, improving data locality for large in‑memory workloads.
- Network and storage bandwidth: Largest sizes reach 100 Gbps network and 72 Gbps EBS throughput, roughly twice the previous ceiling.
- Packet processing: Reported up to three‑fold increase, which can reduce latency for high‑throughput services.
- Instance Bandwidth Configuration (
IBC): Allows a 25% shift of bandwidth between Amazon VPC networking and EBS, giving finer control for database or cache workloads.
Why the change matters to AI, cloud, and DevOps teams
Memory‑intensive workloads such as large language model inference, in‑memory caches (Redis, Valkey, Memcached), and real‑time analytics can now run on fewer vCPUs while achieving higher throughput. The faster DDR5 memory and larger cache reduce latency for data‑heavy pipelines, which translates into lower cost per inference or query when the same workload is migrated from R8g to R9g.
For containerized micro‑services platforms (EKS, ECS, Kubernetes) the higher network ceiling and configurable IBC help avoid bottlenecks when services combine heavy compute with large EBS volumes. The R9gd variant adds local NVMe SSD storage, useful for workloads that need sub‑millisecond storage access without sacrificing the memory‑optimized profile.
Operational and security implications
All R9g/R9gd instances continue to run on the AWS Nitro System, which offloads virtualization, storage, and networking to dedicated hardware. The Nitro Isolation Engine, formally verified as part of the Nitro hypervisor, provides mathematically proven isolation between VMs. While the verification does not eliminate all risk, it raises the assurance level for multi‑tenant environments and may simplify compliance arguments for regulated workloads.
From an operations perspective, the larger instance sizes (up to metal-48xl) expose higher bandwidth limits, so monitoring and capacity‑planning tools should be updated to capture the new ceiling values. The ability to re‑balance bandwidth with IBC means that SREs can tune performance without resizing instances, but they must also track the resulting network/EBS split to avoid unintended throttling.
Related CloudNinjas coverage: AWS.
What This Means For Practitioners
- Evaluate existing R8g workloads for a potential move to R9g/R9gd to gain compute and bandwidth headroom without increasing instance count.
- Leverage
IBCto fine‑tune bandwidth allocation for database or cache services that are EBS‑bound versus network‑bound. - Update monitoring dashboards to reflect the new maximum network (100 Gbps) and EBS (72 Gbps) limits, and verify that auto‑scaling policies consider the higher throughput.
- Consider the Nitro Isolation Engine’s formal verification as a factor when designing multi‑tenant or compliance‑sensitive environments.
- Plan for the local NVMe option in R9gd when low‑latency storage is a requirement, but remember that it replaces EBS‑only storage for those sizes.