The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Short answer: AWS Graviton2 can deliver better price performance than comparable-generation x86 EC2 instances for some workloads, but AWS’s “up to 40%” figure is not a universal speed or cost guarantee. The practical choice is between specific instance types—such as Graviton2-based M6g and Intel-based M5—not between Arm and Intel in the abstract. Benchmark your application and check its Arm64 dependencies before migrating.
What is being compared?
AWS Graviton2 is a 64-bit Arm processor based on Arm Neoverse N1 cores. It powers several Amazon EC2 families, including general-purpose M6g, compute-optimized C6g, memory-optimized R6g, and burstable T4g. The x86 side is not one fixed processor: AWS’s comparisons include particular x86-64 instance families and generations, with Intel-based options such as M5 and M6i. AWS identifies M6g as Graviton2-based and M6i as Intel Xeon Ice Lake-based in its general-purpose instance specifications.
As an Amazon Associate I earn from qualifying purchases.
That distinction matters when choosing a benchmark. Compare the exact instance type and generation you can deploy, not “Arm” against “Intel” as if each were a single configuration. Instance family, size, memory, software stack, Region, and purchase model can all affect the result.
How should you interpret AWS’s performance claim?
AWS says Graviton2 instances can provide “up to 40% better price performance” than specified comparable-generation x86 instances. The qualification matters: it is AWS’s claim for relevant family comparisons and workloads, not a promise that every application will run 40% faster or cost 40% less. AWS’s R6g material, for example, frames its comparison against R5; its M6g comparisons use M5 as the counterpart. See AWS’s R6g instance information and its Graviton performance-testing guidance.
#1 Best Overall
Price performance combines the work an instance performs with what it costs. A lower hourly rate alone does not establish better value if an application needs more instances, misses its latency target, or requires significant porting work. Conversely, an application that handles its workload efficiently on Graviton2 may be able to meet its service targets with less compute. Measure the outcome that matters to your service.
Why equal vCPU counts may not mean equal resources
AWS’s documented comparison distinguishes the core mappings: an x86 vCPU in the cited comparison represents a logical core using simultaneous multithreading (SMT), while a Graviton vCPU represents a physical core; Graviton has no SMT. As a result, matching two instance types by vCPU count does not mean they have the same number of physical cores or guarantee equivalent performance. Check the instance specifications and test the application rather than assuming that equal vCPU labels make a like-for-like comparison. AWS explains its approach in the EC2 Graviton Getting Started guide.
What do the published benchmarks show?
AWS’s .NET 5 test
In a December 2020 AWS Compute Blog benchmark, ASP.NET Core ran under load on m5.xlarge and m6g.xlarge instances. Each configuration had four vCPUs and 16 GiB of memory. The post reported historical hourly prices in us-east-2 (Ohio) of $0.192 for m5.xlarge and $0.154 for m6g.xlarge at that time. Those rates are not current quotes; the results apply to the test’s particular .NET version, application, and method. Read the AWS benchmark post for its setup and findings.
Dream11’s reported results
An AWS-hosted case study reports Dream11’s results for its own workload. In one comparison, 80 m5.4xlarge instances recorded 243 ms average latency and 42% aggregate CPU utilization; 80 m6g.4xlarge instances recorded 204 ms average latency and 32% aggregate CPU utilization. The case study also says 60 M6g instances achieved nearly the same PDF generation time as 80 M5 instances. These are results reported for Dream11’s tests, not evidence that another application will see the same latency or instance reduction. The case study describes the workload and context.
How to compare candidate instances for your workload
AWS recommends establishing a baseline and fully loading both instance types to find the maximum sustainable load before latency or error rates exceed acceptable limits. Apply that guidance to the same production-representative application, data, and traffic profile on each candidate.
- Choose a fair pair. Select the specific Graviton2 and x86 instance types you would actually run, and record their vCPU, memory, family, and generation. Avoid treating vCPU count alone as proof of equivalence.
- Set acceptance limits first. Write down the throughput you need and the latency and error-rate thresholds your service must meet.
- Run the same workload on both. Use comparable application builds, data, and traffic patterns, and load each candidate until it reaches its sustainable limit.
- Capture the whole operating picture. Record throughput, average and tail latency, error rates, CPU and memory use, and the number of instances required to meet your target.
- Compare cost for the same Region and purchase model. Use current prices for your intended deployment; historical benchmark rates are not a substitute for a current quote.
This approach separates a useful result—meeting the same service target at a lower total compute cost—from a misleading comparison based on one headline score or hourly rate. AWS’s getting-started guide provides its baseline and load-testing recommendations.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Check Arm64 compatibility before migration
Performance is only part of the decision. Verify the full software stack for the specific Arm64 versions you plan to use:
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
- Operating system and its release
- Language runtime and application dependencies
- Native libraries, extensions, and build tooling
- Container base images and any binaries they include
- Monitoring, security, backup, and deployment agents
- Third-party products and vendor support for your exact configuration
AWS lists Amazon Linux 2, Red Hat, SUSE, and Ubuntu among popular supported distributions, along with a range of software and services. That list does not establish compatibility for every version or deployment. Check the specific vendor documentation and test the complete application before changing production capacity. AWS’s overview is available in its Graviton guide.
Quick Recap
Best Value
When does a Graviton2 trial make sense?
- Consider testing M6g, C6g, R6g, or T4g when the workload matches the family’s purpose, your dependencies support Arm64, and a controlled benchmark can verify service targets and total cost.
- Keep the current x86 choice for now if a required dependency has no supported Arm64 build, migration risk is unacceptable, or representative testing does not meet your latency, error-rate, or throughput targets.
- Reassess the comparison over time if you are evaluating a new deployment: instance catalogs and prices change, so a historical M5-versus-M6g result cannot establish the value of every current x86 alternative.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
