Amazon Web Services Inc.‘s next-generation silicon is working its way deeper into the company’s cloud compute infrastructure offerings with the debut of the new Amazon EC2 C9g and EC2 C9gd instances today.

The new virtual machines are powered by the company’s flagship AWS Graviton5 processors, which are central processing units that have been designed to deliver superior compute performance for diverse cloud workloads, including artificial intelligence applications.

The C9g and C9gd instances debut just weeks after AWS launched its general-purpose M9g and M9gd instances, and make the new CPUs available for a range of higher-performance workloads.

In a blog post, AWS Principal Developer Advocate Sébastien Stormacq said the C9g and C9gd instances are designed for compute-intensive and data-heavy tasks including video encoding, scientific modeling and distributed analytics, where performance and efficiency are vital. He explained that the Graviton5 chips provide a major leap forward in this regard, with up to 25% higher performance per vCPU compared to the previous-generation C8g instances that were powered by Graviton4.

The advantage of Graviton5 is that it delivers a massive increase in memory and cache, Stormacq said. Both C9g and C9gd feature DDR5 8800MT/s DIMMs, which is the fastest memory of any cloud-based instance currently available. They also pack in five times more L3 cache and three times higher packet-processing performance than Graviton4, dramatically reducing the time workloads sit idle while waiting for the data they need.

According to Stormacq, this architecture means the C9g and C9gd instances are perfectly suited for next-generation “agentic AI” workloads, where artificial intelligence models complete complex, multistep tasks on behalf of humans. Such workloads increase the computational strain on the underlying processors, which is why Graviton5’s higher core count and bigger caches become necessary.

The C9g instances are optimized for workloads that employ the Amazon Elastic Block Store for storage, including video encoding pipelines and batch processing jobs. Meanwhile, the C9gd instances are designed for applications that rely on high-speed and low-latency local NVMe SSD storage, Stormacq said. In this case, they deliver up to 30% higher throughput and input/output operations compared to the previous-generation instances designed for local storage. They’re a great fit for workloads such as temporary caches, ad-serving engines and high-performance computing simulations, he added.

Customers have plenty of options for scaling up the new instances, with 11 sizes available. These range from the medium-sized VMs that feature one virtual CPU and 2 gigabytes of memory, all the way up to 48xlarge, which pack 192 vCPS and 384 gigabytes of memory.

There are also bare-metal server options, and there are massive gains in network and Elastic Block Store throughput, which have increased by 15% and 20% respectively across instance sizes. In the case of the largest 48xlarge instances, these bring 100 gigabytes per second of network bandwidth and 72 gigabits per second of EBS bandwidth to the table, which is almost twice the previous limits.

Stormacq also discussed a critical new security feature that comes with the instances. The C9g and C9gd families are the first from the company to support the new AWS Nitro Isolation Engine, which is designed to enforce isolation between virtual machines. It works by restricting access to the VM’s memory, CPU states and I/O devices to a minimal set of application programming interfaces to guarantee data confidentiality and integrity.

The new C9g and C9gd instances are generally available now in the US East (Ohio, N. Virginia), US West (Oregon), and Europe (Frankfurt) regions, with additional regions to follow later in the year. They’re accessible via the AWS Management Console, the Command Line Interface and also AWS’ software development kits.

Photo: SiliconANGLE