Cuong Nguyen·2h agoWhy one zone of our GPU inference fleet worked 50% harder (and the one-line fix)How Kubernetes topology-aware routing and a “soft” topology spread quietly skewed load across our Triton Inference Server pods, and how we…
