Scheduling GPU Workloads in Kubernetes: From Standard Mechanisms to Custom Solutions

Immerse yourself in the exciting world of GPU and not only GPU scheduling in Kubernetes, where the standard kube-scheduler faces unique challenges in high-performance computing.

We will explore the full range of features — from embedded scheduling mechanisms and their customization to advanced specialized solutions such as Volcano, Apache YuniKorn and KAI Scheduler. You will see how to increase the efficiency of using expensive GPU resources and reduce latency for mission-critical workloads.

Regardless of the scale of your project and its specifics — from small Kubernetes clusters with no GPU at all to artificial intelligence production systems — you will receive valuable tools for gradual implementation: starting with simple basic settings, customization of the standard scheduler, to combining several approaches if necessary.

Join us to discover when and why it's worth switching to custom shadowing solutions, and when the standard ones will be enough.