Posted On: Sep 16, 2022
AWS Fargate customers can now configure Amazon Elastic Container Service (ECS) tasks and Amazon Elastic Kubernetes Service (EKS) pods to use up to 16 vCPUs, an approximately 4x increase from before. vCPUs are the primary compute resource in ECS tasks and EKS pods. Larger vCPUs enable compute-heavy applications like machine learning inference, scientific modeling, and distributed analytics to more easily run on Fargate. In addition, customers can now provision up to 120 GiB of memory on Fargate, also a 4x increase from before. This helps batch workloads, extract, transform, load (ETL) jobs, genomics, and media processing applications better perform memory-intensive operations on Fargate. Larger vCPU and memory options may also make migration to serverless container compute simpler for applications that need more compute resources and cannot be easily re-architected into smaller microservices.
To run container applications on Fargate with these increased vCPU and memory configurations, simply create a new task definition or pod specification with the new requirements, or update existing ones and relaunch. Edit the ECS Task memory and Task CPU parameter values for the Fargate launch type, or update the vCPU and memory reservations for EKS pods that run on Fargate to use higher vCPU and memory values on Fargate. Fargate On-Demand, the Fargate Spot capacity provider, and Savings Plans pricing options automatically extend to these increased vCPU and memory configurations. Customers on ECS can use increased tasks on Fargate for their Linux/X86 and Linux/ARM CPU architectures.
Fargate pricing continues to be based on requested vCPU and memory resources required for the task or pod. See Fargate pricing for more details. Increased vCPU and memory configurations requires the use of Fargate’s vCPU-based Service Quotas, learn more about vCPU-based Service Quotas on the FAQ page. Increased vCPU and memory configurations are available in every commercial AWS Region where Fargate is available as of August 31, 2022, excluding AWS China Regions and AWS GovCloud (US) Regions.