KubeCon + CloudNativeCon India 2026

Running GPU inference on Kubernetes is no longer exotic — it’s becoming the default for modern AI workloads. But while teams obsess over model latency and throughput, the real problems usually hide deeper: GPU under-utilization, memory fragmentation, node-level contention, noisy neighbour, and observability gaps that make debugging feel like guesswork.
In this talk, we’ll walk through a practical, field-tested monitoring approach for GPU inference workloads on Kubernetes. Attendees will learn how to instrument GPU nodes, collect and correlate GPU-specific metrics, build alerting around inference SLOs, and detect performance regressions before they disrupt production. We’ll also cover common anti-patterns and what “good” looks like for GPU observability in 2025.
If you're running (or planning to run) GPU inference at scale, this session will help you monitor responsibly — and keep your cluster healthy, efficient, and fast.