Ignite

Managing Cloud Costs with Kubecost: InVideo’s Journey to Efficient GPU and EKS Operations” At InVideo, we’re revolutionizing video creation through AI, enabling users to generate and edit professional videos with just a prompt. However, running AI workloads on AWS EKS clusters with GPU resources presents significant cost management challenges. How do you effectively track costs across multiple environments and applications? How do you optimize GPU usage between spot and on-demand instances? In this talk, we’ll share InVideo’s journey of implementing Kubecost to tackle these challenges head-on.

You’ll learn how we transformed our cost management strategy, going beyond basic cloud cost reports to gain granular insights into application-level expenses and resource utilization. We’ll demonstrate how Kubecost helped us optimize our EKS clusters and GPU NodeGroups, providing detailed visibility into costs, network usage, and cluster performance. A key highlight will be our implementation of dynamic switching between spot and on-demand instances using Karpenter, significantly reducing our GPU computing costs while maintaining reliability.

The presentation includes a practical demo of Kubecost installation and AWS integration, making it valuable for both beginners and experienced cloud architects. We’ll share real metrics showing the cost savings achieved and discuss lessons learned during implementation. Whether you’re running AI workloads, managing multiple Kubernetes environments, or simply looking to optimize your cloud spending, this talk will provide actionable insights for implementing effective cost management strategies in your organization.