View a markdown version of this page

GAMESUS02-BP02 Right-size your compute and deploy GPU performance only where needed - Games Industry Lens
Implementation guidance

GAMESUS02-BP02 Right-size your compute and deploy GPU performance only where needed

Architect your game servers and backend to efficiently utilize compute resources. Over-provisioning compute can lead to unnecessary costs and minimizes the amount of idle or under-utilized resources. GPU instances should be used to support specific development efforts like HLOD rebuilds in Unreal, or if you game servers require them by design. This greatly reduces the environmental impact and costs of your workloads.

Level of risk exposed if this best practice is not established: High

Implementation guidance

You should optimize your game servers and backend services to use multiple EC2 instances types and the least number of instances needed. This increases the number of instances available to meet your needs during development or for your games launch. You should also match the instance type to the specific workload you're deploying. Compute Optimized instances support a wide range of use-cases including game servers and backend services like matchmaking. Memory Optimized instances are designed to deliver fast performance for workloads that process large data sets in memory. Use GPU instances as required for high performance requirements, but not for general computing tasks. If able, architect your services or game servers to run on ARM with AWS Graviton instances. Graviton is the most performant to energy efficient instance type available on AWS. They also offer improved performance and costs when compared to x86 instance types.

Use the AWS Compute Optimizer to identify the optimal AWS resource configurations, such as Amazon Elastic Compute Cloud (EC2) instance types, Amazon Elastic Block Store (EBS) volume configurations, task sizes of Amazon Elastic Container Service (ECS) services on AWS Fargate, commercial software licenses, AWS Lambda function memory sizes, and Amazon Relational Database Service (RDS) DB instance classes, using machine learning to analyze historical utilization metrics. Compute Optimizer provides a set of APIs and a console experience to reduce costs and increase workload performance by recommending the optimal AWS resources for your AWS workloads.

Implementation steps

  • Match compute resources to specific workloads by using Compute Optimized instances for game servers, Memory Optimized instances for large data sets, and GPU instances only for tasks like HLOD rebuilds or GPU-dependent game servers.

  • Optimize compute utilization by deploying AWS Graviton instances where possible for energy efficiency, better performance, and cost savings compared to x86 instances.

  • Use AWS Compute Optimizer to analyze historical utilization and recommend the most efficient configurations for EC2, AWS ECS, AWS Lambda, and Amazon RDS workloads to reduce costs and improve performance.