Amazon CloudWatch Container Insights adds Sub-Minute GPU Metrics for Amazon EKS

Amazon CloudWatch Container Insights

Now supports collection of GPU metrics at sub-minute frequencies for AI and ML workloads running on Amazon EKS. Customers can configure the metric sample rate in seconds, enabling more granular monitoring of GPU resource utilization.

This enhancement enables customers to effectively monitor GPU-intensive workloads that run for less than 60 seconds, such as ML inference jobs that consume GPU resources for short durations. By increasing the sampling frequency, customers can maintain detailed visibility into short-lived GPU workloads. Sub-minute GPU metrics are sent to CloudWatch once per minute. This granular monitoring helps customers optimize their GPU resource utilization, troubleshoot performance issues, and ensure efficient operation of their containerized GPU applications.

Sub-Minute GPU metrics in Container Insights is available in all AWS Commercial Regions and the AWS GovCloud (US) Regions.

What to do

• Configure the metric sample rate in seconds for your GPU workloads.
• Monitor GPU resource utilization more granularly.
• Optimize GPU resource utilization and troubleshoot performance issues.

Source: AWS release notes

If you need further guidance on AWS, our experts are available at AWS@westloop.io. You may also reach us by submitting the Contact Us form.