Amazon Web Services opened a preview of its Graviton2-powered M6g cloud instances on December 3, 2019, giving customers an Arm-based option to evaluate for more demanding applications. AWS restricted the preview to non-production workloads. For a business considering a move from its existing EC2 instances, that made the announcement an opportunity to test an alternative—not evidence that a production migration would already save money. [S1][S2]

AWS put a number on the attraction: it claimed up to 40% better price/performance for the new Graviton2-based M, R and C families than for its x86-based M5, R5 and C5 instances. Price/performance relates the price paid to the work delivered; it is not simply the hourly rental price. The claim therefore cannot be read as 40% off every customer’s bill. A buyer needs to compare the resources and cost required to deliver the same application workload at an acceptable service level. [S2]

The initial benchmarks in AWS’s accompanying technical post show why the workload matters. Against M5 instances, AWS reported per-virtual-CPU performance gains of 24% for HTTPS load balancing with Nginx, 43% for Memcached, 26% for X.264 video encoding and 54% for a Cadence Xcellium simulation. These were AWS’s tests, not independent measurements, and they were performance figures rather than a table of total application costs. None establishes the result for a different company’s software or traffic. [S1]

Software readiness is part of that comparison. Graviton2 uses 64-bit Arm Neoverse cores, whereas the M5, R5 and C5 comparison families use x86 processors. AWS named support from Linux distributions, its Corretto distribution of OpenJDK, container services, monitoring agents and developer tools. That ecosystem gives a buyer starting points to check; it does not certify every application, dependency or proprietary component. Our purchasing test would include those dependencies and the engineering effort needed to prepare and support the application, rather than treating a supported operating system as a completed migration. [S2]

Availability also had separate stages. M6g was in preview, while AWS said C6g, C6gd, M6gd, R6g and R6gd would follow in the coming months. The company said several of its own services had tested Graviton2 and planned production use in 2020. Those were plans, not completed deployments on December 3. Neither announcement supplies a basis for assuming that every instance type was ready for production in every AWS region. [S1][S2]

For a buyer, the useful comparison would hold the application’s output and response-time requirements steady, then record the compute resources, other service charges and migration work needed on each option. This is an evaluation method, not a saving measured by these announcements. If extra capacity or engineering work absorbs the compute advantage, the lower-cost case weakens; if the application meets the same requirements with lower overall cost, it strengthens. Both source documents come from AWS, so the deciding evidence would be the buyer’s own matched workload test—not a second reading of the same supplier’s claim. [S1][S2]

Source documents

  1. Coming Soon – Graviton2-Powered General Purpose, Compute-Optimized, & Memory-Optimized EC2 Instanceshttps://aws.amazon.com/blogs/aws/coming-soon-graviton2-powered-general-purpose-compute-optimized-memory-optimized-ec2-instances/
  2. AWS Announces Nine New Compute and Networking Innovations for Amazon EC2https://press.aboutamazon.com/2019/12/aws-announces-nine-new-compute-and-networking-innovations-for-amazon-ec2