NVIDIA 开源 Topograph：面向 AI 工厂的拓扑感知工作负载调度
原标题：Topology-Aware Workload Scheduling with NVIDIA Topograph
AI 摘要
NVIDIA 发布开源工具包 Topograph，用于发现集群网络拓扑并将其标准化为 Kubernetes 节点标签、Slurm 配置、Slinky ConfigMap 等调度器可识别的格式。它通过 API Server、Node Observer、Node Data Broker、Provider 和 Engine 五个组件保持拓扑视图实时更新，并与 DRA、KAI Scheduler 配合实现拓扑感知的 gang 调度。该工具支持 Google Cloud、Lambda、Nebius、Nscale、OCI 等云厂商及 InfiniBand、Spectrum-X 本地环境，旨在减少跨域通信瓶颈、提升 GPU 利用率并降低作业成本。
正文节选
AI factories are power-limited systems that deliver maximum value when fully optimized. GPU workload placement is a key optimization. Poor workload placement fragments topology domains and forces traffic across shared links, reducing throughput, raising job costs, and leaving GPUs consuming provisioned power while waiting on data without advancing the workload. GPUs exchange data continuously during training and inference, so distributed workloads benefit from communication locality. NVIDIA NVLi