# Hardware patterns
**Topics**
+ [SUS 5 How do your hardware management and usage practices support your sustainability goals?](sus-05.md)
# SUS 5 How do your hardware management and usage practices support your sustainability goals?
Look for opportunities to reduce workload sustainability impacts by making changes to your hardware management practices. Minimize the amount of hardware needed to provision and deploy, and select the most efficient hardware for your individual workload.
**Topics**
+ [SUS05-BP01 Use the minimum amount of hardware to meet your needs](sus_sus_hardware_a2.md)
+ [SUS05-BP02 Use instance types with the least impact](sus_sus_hardware_a3.md)
+ [SUS05-BP03 Use managed services](sus_sus_hardware_a4.md)
+ [SUS05-BP04 Optimize your use of GPUs](sus_sus_hardware_a5.md)
# SUS05-BP01 Use the minimum amount of hardware to meet your needs
Using the capabilities of the cloud, you can make frequent changes to your workload implementations. Update deployed components as your needs change.
**Level of risk exposed if this best practice is not established:** Medium
## Implementation guidance
+ Enable horizontal scaling, and use automation to scale out as loads increase and to scale in as loads decrease.
+ Scale using small increments for variable workloads.
+ Align scaling with cyclical utilization patterns (for example, a payroll system with intense bi-weekly processing activities) as load varies over days, weeks, months, or years.
+ Negotiate service level Agreements (SLAs) that allow for a temporary reduction in capacity while automation deploys replacement resources.
## Resources
**Related documents:**
+ [AWS Compute Optimizer Documentation](https://docs.aws.amazon.com/compute-optimizer/index.html)
+ [Operating Lambda: Performance optimization](https://aws.amazon.com/blogs/compute/operating-lambda-performance-optimization-part-2/)
+ [Auto Scaling Documentation](https://docs.aws.amazon.com/autoscaling/index.html)
# SUS05-BP02 Use instance types with the least impact
Continually monitor the release of new instance types and take advantage of energy efficiency improvements, including those instance types designed to support specific workloads such as machine learning training, inference, and video transcoding.
**Common anti-patterns:**
+ You are only using one family of instances.
+ You are only using x86 instances.
+ You specify one instance type in your Amazon EC2 Auto Scaling configuration.
+ You use AWS instances in a manner that they were not designed for (for example, you use compute-optimized instances for a memory-intensive workload).
+ You do not evaluate new instance types regularly.
+ You do not check recommendations from AWS rightsizing tools such as [AWS Compute Optimizer.](https://aws.amazon.com/compute-optimizer/)
**Benefits of establishing this best practice:** By using energy-efficient and right-sized instances, you are able to greatly reduce the environmental impact and cost of your workload.
**Level of risk exposed if this best practice is not established:** Low
## Implementation guidance
+ Learn and explore instance types which can lower your workload environmental impact.
+ Subscribe to [What's New with AWS](https://aws.amazon.com/new/) to be up-to-date with the latest AWS technologies and instances.
+ Learn about different AWS instance types.
+ Learn about AWS Graviton-based instances which offer the best performance per watt of energy use in Amazon EC2 by watching [re:Invent 2020 - Deep dive on AWS Graviton2 processor-powered Amazon EC2 instances](https://www.youtube.com/watch?v=NLysl0QvqXU) and [Deep dive into AWS Graviton3 and Amazon EC2 C7g instances](https://www.youtube.com/watch?v=WDKwwFQKfSI&ab_channel=AWSEvents).
+ Plan and transition your workload to instance types with the least impact.
+ Define a process to evaluate new features or instances for your workload. Take advantage of agility in the cloud to quickly test how new instance types can improve your workload environmental sustainability. Use proxy metrics to measure how many resources it takes you to complete a unit of work.
+ If possible, modify your workload to work with different numbers of vCPUs and different amounts of memory to maximize your choice of instance type.
+ Consider transitioning your workload to Graviton-based instances to improve the performance efficiency of your workload (see [AWS Graviton Fast Start](https://aws.amazon.com/ec2/graviton/fast-start/) and [AWS Graviton2 for ISVs](https://docs.aws.amazon.com/whitepapers/latest/aws-graviton2-for-isv/welcome.html)). Keep in mind the [considerations when transitioning workloads to AWS Graviton-based Amazon Elastic Compute Cloud instances.](https://github.com/aws/aws-graviton-getting-started/blob/main/transition-guide.md)
+ Consider selecting the AWS Graviton option in your usage of [AWS managed services.](https://github.com/aws/aws-graviton-getting-started/blob/main/managed_services.md)
+ Migrate your workload to Regions that offer instances with the least sustainability impact and still meet your business requirements.
+ For machine learning workloads, use Amazon EC2 instances which are based on custom Amazon Machine Learning chips such as [AWS Trainium](https://aws.amazon.com/machine-learning/trainium/), [AWS Inferentia](https://aws.amazon.com/machine-learning/inferentia/), and [Amazon EC2 DL1.](https://aws.amazon.com/ec2/instance-types/dl1/)
+ Use [Amazon SageMaker AI Inference Recommender](https://docs.aws.amazon.com/sagemaker/latest/dg/inference-recommender.html) to right size ML inference endpoint.
+ For workloads with real time video transcoding, use [Amazon EC2 VT1 Instances.](https://aws.amazon.com/ec2/instance-types/vt1/)
+ For spikey workloads (workloads with infrequent requirements for additional capacity), use [burstable performance instances.](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/burstable-performance-instances.html)
+ For stateless and fault-tolerant workloads, use [Amazon EC2 Spot Instances](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/using-spot-instances.html) to increase overall utilization of the cloud, and reduce the sustainability impact of unused resources.
+ Operate and optimize your workload instance.
+ For ephemeral workloads, evaluate [instance Amazon CloudWatch metrics](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/viewing_metrics_with_cloudwatch.html#ec2-cloudwatch-metrics) such as `CPUUtilization` to identify if the instance is idle or under-utilized.
+ For stable workloads, check AWS rightsizing tools such as [AWS Compute Optimizer](https://aws.amazon.com/compute-optimizer/) at regular intervals to identify opportunities to optimize and right-size the instances.
## Resources
**Related documents:**
+ [Optimizing your AWS Infrastructure for Sustainability, Part I: Compute](https://aws.amazon.com/blogs/architecture/optimizing-your-aws-infrastructure-for-sustainability-part-i-compute/)
+ [AWS Graviton Processor](https://aws.amazon.com/ec2/graviton/)
+ [AWS Inferentia](https://aws.amazon.com/machine-learning/inferentia/)
+ [AWS Trainium](https://aws.amazon.com/machine-learning/trainium/)
+ [Amazon EC2 DL1](https://aws.amazon.com/ec2/instance-types/dl1/)
+ [Amazon EC2 Burstable performance instances](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/burstable-performance-instances.html)
+ [Amazon EC2 Capacity Reservation Fleets](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/cr-fleets.html)
+ [Amazon EC2 Spot Fleet](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/spot-fleet.html)
+ [Amazon EC2 Spot Instances](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/using-spot-instances.html)
+ [Amazon EC2 VT1 Instances](https://aws.amazon.com/ec2/instance-types/vt1/)
+ [Amazon EC2 instance types](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/instance-types.html)
+ [AWS Compute Optimizer](https://aws.amazon.com/compute-optimizer/)
+ [Functions: Lambda Function Configuration](https://docs.aws.amazon.com/lambda/latest/dg/best-practices.html#function-configuration)
**Related videos:**
+ [Deep dive on AWS Graviton2 processer-powered Amazon EC2 instances](https://www.youtube.com/watch?v=NLysl0QvqXU)
+ [Deep dive into AWS Graviton3 and Amazon EC2 C7g instances](https://www.youtube.com/watch?v=WDKwwFQKfSI&ab_channel=AWSEvents)
**Related examples:**
+ [Lab: Rightsizing Recommendations](https://wellarchitectedlabs.com/cost/100_labs/100_aws_resource_optimization/)
+ [Lab: Rightsizing with Compute Optimizer](https://wellarchitectedlabs.com/cost/200_labs/200_aws_resource_optimization/)
+ [Lab: Optimize Hardware Patterns and Observe Sustainability KPIs](https://wellarchitectedlabs.com/sustainability/200_labs/200_optimize_hardware_patterns_observe_sustainability_kpis/)
# SUS05-BP03 Use managed services
Managed services shift responsibility for maintaining high-average utilization, and sustainability optimization of the deployed hardware to AWS. Use managed services to distribute the sustainability impact of the service across all tenants of the service, reducing your individual contribution.
**Level of risk exposed if this best practice is not established:** Low
## Implementation guidance
+ Migrate from self-hosted services to managed services. For example, use managed [Amazon Relational Database Service (Amazon RDS)](https://aws.amazon.com/rds/) instances instead of maintaining your own Amazon RDS instances on [Amazon Elastic Compute Cloud (Amazon EC2)](https://aws.amazon.com/ec2/), or use managed container services, such as [AWS Fargate](https://aws.amazon.com/fargate/), instead of implementing your own container infrastructure.
## Resources
**Related documents:**
+ [AWS Fargate](https://aws.amazon.com/fargate/)
+ [Amazon DocumentDB](https://aws.amazon.com/documentdb/)
+ [Amazon Elastic Kubernetes Service (EKS)](https://aws.amazon.com/eks/)
+ [Amazon Managed Streaming for Apache Kafka (Amazon MSK)](https://aws.amazon.com/msk/)
+ [Amazon Redshift](https://aws.amazon.com/redshift/)
+ [Amazon Relational Database Service (RDS)](https://aws.amazon.com/rds/)
# SUS05-BP04 Optimize your use of GPUs
Graphics Processing Units (GPUs) can be a source of high-power consumption, and many GPU workloads are highly variable, such as rendering, transcoding, and machine learning training and modeling. Only run GPU instances for the time needed, and decommission them with automation when not required to minimize resources consumed.
**Level of risk exposed if this best practice is not established:** Low
## Implementation guidance
+ Use GPUs only for tasks where they’re more efficient than CPU-based alternatives.
+ Use automation to release GPU instances when not in use.
+ Use flexible graphics acceleration rather than dedicated GPU instances.
+ Take advantage of custom-purpose hardware that is specific to your workload.
## Resources
**Related documents:**
+ [Accelerated Computing](https://aws.amazon.com/ec2/instance-types/#Accelerated_Computing)
+ [AWS Inferentia](https://aws.amazon.com/machine-learning/inferentia/)
+ [AWS Trainium](https://aws.amazon.com/machine-learning/trainium/)
+ [Accelerated Computing for EC2 Instances](https://aws.amazon.com/ec2/instance-types/#Accelerated_Computing)
+ [Amazon EC2 VT1 Instances](https://aws.amazon.com/ec2/instance-types/vt1/)
+ [Amazon Elastic Graphics](https://docs.aws.amazon.com/AWSEC2/latest/WindowsGuide/elastic-graphics.html)