FSISUS14: Do you have multi-architecture images for grid computing systems?
Multi-architecture image support for a particular workload makes it easier for you to build different images and thus different architectures and operating systems from the same source and refer to them all by the same abstract manifest. The manifest specifies the layers of system content that make up the image as well as its runtime characteristics and configuration. Having a multi-architecture image increases the flexibility of the workload thus increases the opportunity to use hardware that may be more sustainable.
FSISUS14-BP01 Use instances with higher energy efficiency
Prescriptive guidance
-
AWS Graviton-based instances use up to 60% less energy than comparable EC2 instances.
FSISUS14-BP02 Design applications that can use different Amazon EC2 instance types
Prescriptive guidance
-
This is what we would call a flexible workload. In contrast, inflexible workloads rely only on a few instance types. These instances types may be less energy efficient than others.
-
Flexible workloads are ideal for Spot Instances. Running workloads on Spot Instances is generally considered more energy efficient than On-Demand Instances because Spot is overhead required for the Amazon EC2 On-Demand service to run.
-
Use Amazon EC2's spare capacity with Spot Instances to extract the same value, which increases the total value generated from the Amazon EC2 environment as a whole.
FSISUS14-BP03 Adopt a serverless, event-driven architecture
Prescriptive guidance
-
Consider using a serverless, event-driven architecture to maximize overall resource utilization. Serverless architecture removes the requirement to run and maintain physical servers since AWS handles this on your behalf.
-
The cost of serverless architectures generally correlates with the level of usage, thus increases your workload's cost efficiency.
-
Implementation guidance: Maximize energy efficiency as well as availability by building multi- architecture workloads that can run on a variety of Spot Instances. It is important to account for error precision when expanding compiler options on varying processors.
-
Service recommendations: Use the following services to achieve your goal:
-
Determine which of your workloads is suitable for use of floating-point accuracy, performance, and efficiency. Consider testing with a cluster of instances to see how well it performs at scale.
-
For intensive financial simulations and calculations, test the number of bits that are required to achieve your floating point precision and consider reducing number of bits by selecting different floating-point formats, including bfloat16, that's supported by AWS Graviton.
-
Develop multi-architecture generative AI model containers for different instance types.
-
Support both GPU and AWS Trainium instances for generative AI workloads.
-
Optimize generative AI models for different hardware architectures (like x86, ARM, or Graviton).
