Amazon Web Services has introduced the EC2 C9g and C9gd instance families featuring its latest Graviton5 processors. These new instances target compute-heavy applications including AI, video encoding, and scientific workloads by delivering significant gains in CPU performance, memory speed, and security isolation capabilities.
- Up to 25% better performance per vCPU with Graviton5 vs. prior generation
- DDR5 memory and larger caches reduce idle compute wait times
- First AWS instances supporting Nitro Isolation Engine for VM security
Infrastructure signal
The new AWS EC2 C9g and C9gd instances mark a substantial step forward by embedding Graviton5 processors into the high-performance compute segment. These CPUs offer significant architectural improvements including five times more L3 cache and DDR5 8800MT/s DIMMs, the fastest memory configuration available on any cloud instance today. This combination drastically lowers latency caused by memory fetch delays, making the instances well-suited for compute and data-intensive workloads like video processing and scientific simulations.
Furthermore, network and storage throughput improvements enhance overall infrastructure efficiency. The largest 48xlarge instance supports up to 100 Gbps network bandwidth and 72 Gbps EBS bandwidth—nearly doubling the previous generation limits. This networking and storage boost enables faster data transfer and scaling capabilities, elevating performance in distributed architectures and I/O-bound workloads. Together, these advances underline AWS’s focus on optimizing cost-performance ratios at scale.
Developer impact
Developers working on AI models that perform complex multistep computations, such as next-generation agentic AI systems, will find the C9g and C9gd instances particularly advantageous due to the new CPUs’ higher core counts and expanded cache sizes. The substantial jump in memory speed and data caching translates directly to reduced idle CPU cycles waiting for data, accelerating workflows and shortening iteration times for compute-heavy tasks.
The availability of 11 different instance sizes allows teams to right-size infrastructure for both small-scale testing and large production workloads without jeopardizing performance consistency. Also, support for NVMe local SSD storage on C9gd instances improves I/O operations by up to 30% compared to previous generations, benefitting developers managing ephemeral caches or high-throughput transactional systems. These instances are accessible through common AWS tooling, seamlessly integrating into existing CI/CD and deployment pipelines.
What teams should watch
Cloud architects and platform teams should consider the new Nitro Isolation Engine feature included in the C9g and C9gd, which enforces strict separation between virtual machines by limiting access to CPU states, memory, and I/O to minimal, secure APIs. This security addition is critical for multi-tenant and regulated environments, improving data confidentiality and integrity without sacrificing performance. Teams managing workloads with sensitive data should evaluate migration plans to leverage these enhanced protections.
Additionally, operational teams should monitor performance and cost benefits from shifting to DDR5 and higher-granularity instance sizes offered by Graviton5-powered VMs. The improved throughput for both network and Elastic Block Store I/O opens opportunities to redesign some storage architectures for higher elasticity and reduced latency. Observability tools will need calibration to factor in new architecture characteristics, especially to fully track CPU cache efficiency and memory bandwidth utilization.