AWS Unveils Its Most Powerful and Efficient CPU Yet: The 192-Core Graviton5

192 Cores, 3 nm Silicon, and a Re-Architected Memory System

As cloud workloads continue to grow in both size and complexity, AWS has positioned Graviton5 as the company’s strongest and most efficient custom CPU to date. Compared to its predecessor, the new design promises up to 25% more compute performance, while maintaining the sustainability and cost advantages that Graviton processors are known for.

At the heart of the chip lies a densely packed 192-core cluster, the highest core count ever offered in Amazon EC2. The compact layout shortens inter-core data paths, reducing communication latency by up to 33% and further improving bandwidth. The transition to TSMC’s 3 nm node also contributes to improved thermal behavior and lower power draw.

One of the major architectural upgrades comes in the form of a substantially expanded 192 MB L3 cache—five times larger than before. Each core now accesses 2.6× more L3 than on Graviton4, keeping frequently used data closer to the processor and significantly cutting wait times. AWS has also refreshed the memory subsystem, pushing speeds to 7200 MT/s today, with support for 8800 MT/s DIMMs currently in development.

Faster Networking and Storage Throughput

Networking and storage I/O also see notable gains. Graviton5-based EC2 instances deliver, on average, 15% higher network performance and 20% higher EBS bandwidth, with the largest instance sizes capable of doubling available network throughput. These improvements translate directly into faster distributed applications, shorter backup windows, and more consistent performance under heavy load.

Although instances run in a single-socket configuration, AWS distinguished engineer Ali Saidi confirmed that a dual-socket design still exists behind the scenes, with both sockets communicating through a shared Nitro smartNIC. In addition, Graviton5 becomes the first server-class CPU to introduce PCIe 6.0, following in the footsteps of Graviton3, which brought PCIe 5.0 to the cloud.

Real-World Performance Gains Already Verified

Several major companies have already validated early Graviton5 gains.

• Adobe reports faster real-time content personalization thanks to higher compute throughput.
• Epic Games notes smoother gameplay experiences during peak hours, attributing this to lower latency and higher bandwidth.
• Formula 1 uses the chip to process telemetry in seconds and deliver live insights to millions of viewers.
• Pinterest says it can support its 500 million monthly active users more cost-effectively on Graviton-based infrastructure.

New EC2 Instances: M9g Available in Preview, C9g and R9g Coming in 2026

AWS has launched its first Graviton5-powered instances—M9g general-purpose EC2 instances—now available in preview. Compute-optimized C9g and memory-optimized R9g families are slated for release in 2026.

Compared to the previous-generation M8g lineup, the new M9g instances deliver up to 25% better performance. While M8g relied on a dual-socket design with two 96-core processors, Graviton5 consolidates all 192 cores into a single socket, simplifying the architecture and improving efficiency.