NVIDIA Open Sourced the Cluster Map Kubernetes Schedulers Never Had
NVIDIA open sourced Topograph, which brings topology-aware GPU scheduling to Kubernetes by publishing live fabric data as node labels.
Independent reporting on Kubernetes, cloud-native infrastructure, virtualization, storage, and security. For engineers and platform teams.
NVIDIA open sourced Topograph, which brings topology-aware GPU scheduling to Kubernetes by publishing live fabric data as node labels.
Sidero Labs sold to Yardi and added Talos Hypervisor, a native hypervisor inside the Kubernetes OS. VMs and containers now share one immutable host.
Google's AX v0.3.0 moved agent state in etcd out of the critical path. Task state now lives in Redis, and the lesson travels further than the code.
AWS shipped the HyperPod Inference Gateway as one EKS add-on and adopted the open routing spec. First-token latency fell up to 98 percent at P99.
Microsoft open sourced TauGrid, one Helm install that runs AI workloads on Kubernetes across GPU nodes. The assembly is the interesting part.
Cilium 1.20 opens the eBPF datapath to cloud provider plugins and turns Gateway API into a real traffic manager.
Kubernetes 1.37 moved gang scheduling and the Workload and PodGroup APIs to Beta. The scheduler is being rebuilt for accelerators.
A containerd CRI vulnerability let a pod creator poison a node's image cache. Three of five patches were critical. Check your runtime version.
The 2026 Gartner Magic Quadrant for Container Management keeps seven Leaders. Hyperscalers lead, with Red Hat and SUSE as the picks outside them.
Kubernetes GPU scheduling is the AI bottleneck. The scheduler was built for pods, not accelerators, and the utilization numbers show it.
Independent analysis of cloud-native infrastructure, Kubernetes and data center economics. No vendor spin.