AWS has recently announced the general availability of the C7gn and the Hpc7g instances, both using the new Graviton3E processors. The C7gn instances are designed for network-intensive workloads while the Hpc7g instances are tailored for high-performance computing ones.
According to the cloud provider, the Graviton3E processors deliver higher memory bandwidth and compute performance than Graviton2 and up to 35% higher vector-instruction performance compared to Graviton3 processors. Leveraging ARM Neoverse technology, all of the new instances rely on the latest DDR5 memory and 5th generation AWS Nitro Cards, providing improved latency and 60% higher packet-per-second (PPS) performance.
Announced in preview last November, the C7gn instances are supposed to deliver the highest network bandwidth and be the best price-performance option for network-intensive workloads on AWS. Jeff Barr, chief evangelist at AWS, writes:
The instances are designed for your most demanding network-intensive workloads (firewalls, virtual routers, load balancers, and so forth), data analytics, and tightly-coupled cluster computing jobs. They are powered by AWS Graviton3E processors and support up to 200 Gbps of network bandwidth.
In the Last Week in AWS newsletter, Corey Quinn argues that very few technical details were shared:
This blog post is a good example of how AWS talks a lot about Graviton without saying very much about it. There's no technical depth into what differentiates Graviton3 from Graviton 3E.
Starting with only 1 vCPU and 2GiB, the C7gn instances are currently available in 8 different instance sizes in only 4 regions: Ohio, Northern Virginia, Oregon, and Ireland.
On the same day, AWS announced the general availability of the Hpc7g, a new instance class for HPC workloads. The cloud provider published two separate articles on the advantages of the Hpc7g instances, the first one on improved performance, benchmarking popular HPC applications like Siemens Simcenter STAR-CCM+ and Ansys Fluent, and the second on the new sizing approach.
The Hpc7g is the first HPC instance to offer multiple sizes, specifically designed to enhance support for commercial software licensed on a per-core basis. Each instance size varies only in terms of the number of physical cores, while the cost remains consistent across all sizes. The instances feature 128 GiB of memory, 200 Gbps of network performance, and 20 Gbps of EBS performance. Karthik Raman, principal application engineer at AWS, explains:
These different sizes provide an easy way for customers to use the Amazon EC2 Optimize CPU options feature on the Hpc7g instances. This enables customers to choose from a range of instance sizes to target maximum performance per instance or maximum performance per core for their HPC workloads.
The Hpc7g instances are currently available only in Northern Virginia.
Some technical details about the different architectures of Graviton2, Graviton3, and Graviton3E are available in the Graviton Technical Guide on GitHub.