en.Wedoany.com Reported - Amazon Web Services (AWS) has launched the C9g compute-optimized instance powered by its self-developed Graviton5 processor, designed for compute-intensive workloads and now generally available.

The C9g instance features Graviton5 CPUs with twice the core count of the previous generation and a 33% reduction in inter-core latency. It is currently available in the US East, US West, and Europe (Frankfurt) regions, with plans to expand to more regions.

This is AWS's latest move in its compute-optimized product line following the launch of EC2 M9g and M9gd instances in June. AWS stated that compared to the previous-generation C8g instances, the C9g instances deliver up to a 25% improvement in per-virtual CPU (vCPU) performance.

In terms of applications, the instance is optimized for agentic workloads. AWS Principal Developer Advocate Sébastien Stormacq noted that faster processing speeds and larger caches reduce the time workloads wait for data, thereby improving memory analysis throughput, accelerating agent loops, and enhancing real-time application responsiveness. As AI shifts from answering questions to executing tasks, running code, and orchestrating multi-step workflows, the demand for CPU computing is growing, and the C9g instance is designed to handle such tasks.

The concurrently launched C9gd instance comes with local NVMe solid-state drives (SSDs), suitable for scenarios requiring low-latency storage, such as high-performance computing (HPC) simulations and ad serving engines. Compared to previous-generation local storage instances, its high-speed, low-latency storage supports higher throughput and input/output operations per second (IOPS).

The C9g and C9gd are the first compute-optimized AWS instances equipped with the Nitro isolation engine. This security component controls access to virtual machine memory, CPU register states, and I/O devices through a minimal API set, providing mathematically verified isolation between virtual machines. The previously launched M9g and M9gd instances also include this feature.

Both instance series offer 11 sizes ranging from medium to 48xlarge, along with a bare metal option. AWS data shows that compared to the previous generation, different sizes achieve an average network bandwidth improvement of 15% and an Elastic Block Store (EBS) bandwidth improvement of 20%.

AWS is continuously expanding its lineup of instances powered by self-developed chips, providing cloud customers with a broader range of underlying hardware options beyond Nvidia. In March, the cloud provider partnered with chip company Cerebras to combine its Trainium servers with the wafer-scale CS-3 system; meanwhile, efforts to optimize self-developed chips for AI ultimately led to Project Rainier, a massive cluster facility built for Anthropic equipped with over 500,000 Trainium2 chips.

AWS's expansion efforts are accompanied by cost adjustments. Starting July 1, the fees for EC2 capacity blocks used for machine learning GPU reserved services will increase by approximately 20%, marking the second price hike in six months.