[{"data":1,"prerenderedAt":258},["ShallowReactive",2],{"story":3},{"name":4,"created_at":5,"published_at":6,"updated_at":7,"id":8,"uuid":9,"content":10,"slug":249,"full_slug":250,"sort_by_date":25,"position":251,"tag_list":252,"is_startpage":159,"parent_id":253,"meta_data":25,"group_id":254,"first_published_at":6,"release_id":25,"lang":255,"path":256,"alternates":257,"default_full_slug":25,"translated_slugs":25},"AI-INFRA 100-03: GPU Cluster Networking","2026-09-17T18:33:44.800Z","2026-09-18T10:49:22.927Z","2026-09-18T10:49:22.941Z",221103820048656,"05723671-6e71-4da1-8343-3092a328e7fd",{"seo":11,"_uid":15,"type":16,"intro":17,"title":18,"duration":13,"component":19,"course_id":20,"technology":21,"description":22,"on_schedule":159,"course_level":160,"hide_sidebar":161,"prerequisites":162,"on_demand_link":165,"lab_requirements":168,"course_objectives":201,"follow_up_courses":204,"who_should_attend":207,"additional_content":236,"on_demand_training":159,"section_below_hero":241,"hide_class_schedule_cta":161,"hide_private_training_cta":159,"section_below_hero_bg_color":13},{"_uid":12,"title":13,"plugin":14,"og_image":13,"og_title":13,"description":13,"twitter_image":13,"twitter_title":13,"og_description":13,"twitter_description":13},"867ef41d-665c-4afb-ab8c-7929189d5cb2","","seo_metatags","8fdb6453-e42e-443a-befd-e555c41d99b5","course","RoCE v2, PFC/ECN, and fabric fault diagnosis","(AI-INFRA 100-03)","training_course","GPU Cluster Networking","cn",{"type":23,"attrs":24,"content":26},"doc",{"backgroundColor":25},null,[27,121,128],{"type":28,"attrs":29,"content":31},"ordered_list",{"order":30},1,[32,58,79,100],{"type":33,"content":34},"list_item",[35,45,50],{"type":36,"attrs":37,"content":38},"paragraph",{"textAlign":25},[39],{"text":40,"type":41,"marks":42},"Why AI fabric is different","text",[43],{"type":44},"bold",{"type":36,"attrs":46,"content":47},{"textAlign":25},[48],{"text":49,"type":41},"Synchronized RDMA traffic against loss-tolerant TCP, why RDMA requires a lossless fabric, and how PFC and ECN give Ethernet that property. NVIDIA NetQ as the telemetry surface for Ethernet and RoCE, with UFM named as its InfiniBand counterpart.",{"type":36,"attrs":51,"content":52},{"textAlign":25},[53],{"text":54,"type":41,"marks":55},"Lab: RDMA host readiness",[56],{"type":57},"italic",{"type":33,"content":59},[60,67,72],{"type":36,"attrs":61,"content":62},{"textAlign":25},[63],{"text":64,"type":41,"marks":65},"Configuring a lossless fabric",[66],{"type":44},{"type":36,"attrs":68,"content":69},{"textAlign":25},[70],{"text":71,"type":41},"RoCE v2 as RDMA semantics over UDP/IP, per-priority pause frames, and ECN with DCQCN as congestion control that works alongside PFC. Lossless configuration is a two-sided contract between switch and NIC, so both sides get read and set. BlueField-3 on the fabric, and the difference between NIC mode and DPU mode.",{"type":36,"attrs":73,"content":74},{"textAlign":25},[75],{"text":76,"type":41,"marks":77},"Lab: healthy fabric baseline",[78],{"type":57},{"type":33,"content":80},[81,88,93],{"type":36,"attrs":82,"content":83},{"textAlign":25},[84],{"text":85,"type":41,"marks":86},"Fabric observability and fault detection",[87],{"type":44},{"type":36,"attrs":89,"content":90},{"textAlign":25},[91],{"text":92,"type":41},"Fleet-level fabric telemetry as an investigation surface, an evidence hierarchy that separates what decides from what merely corroborates, and the counter signatures that distinguish one fault class from another.",{"type":36,"attrs":94,"content":95},{"textAlign":25},[96],{"text":97,"type":41,"marks":98},"Lab: incident A, blind diagnosis",[99],{"type":57},{"type":33,"content":101},[102,109,114],{"type":36,"attrs":103,"content":104},{"textAlign":25},[105],{"text":106,"type":41,"marks":107},"Diagnosing and fixing misconfiguration",[108],{"type":44},{"type":36,"attrs":110,"content":111},{"textAlign":25},[112],{"text":113,"type":41},"A four-layer recovery check run against a live fault: localize, diagnose, remediate, verify. Configuration drift as the failure mode that scales, and how the same workflow holds at several thousand ports.",{"type":36,"attrs":115,"content":116},{"textAlign":25},[117],{"text":118,"type":41,"marks":119},"Lab: incident B, diagnose and fix. Optional lab: MTU mismatch",[120],{"type":57},{"type":36,"attrs":122,"content":123},{"textAlign":25},[124],{"text":125,"type":41,"marks":126},"FORMAT",[127],{"type":44},{"type":129,"content":130},"bullet_list",[131,138,145,152],{"type":33,"content":132},[133],{"type":36,"attrs":134,"content":135},{"textAlign":25},[136],{"text":137,"type":41},"2 hours, instructor-led",{"type":33,"content":139},[140],{"type":36,"attrs":141,"content":142},{"textAlign":25},[143],{"text":144,"type":41},"Short briefing, then terminal time",{"type":33,"content":146},[147],{"type":36,"attrs":148,"content":149},{"textAlign":25},[150],{"text":151,"type":41},"Two labs are blind diagnoses",{"type":33,"content":153},[154],{"type":36,"attrs":155,"content":156},{"textAlign":25},[157],{"text":158,"type":41},"Hosted lab environment, nothing to install",false,"essentials",true,{"type":23,"content":163},[164],{"type":36},{"id":13,"url":13,"linktype":166,"fieldtype":167,"cached_url":13},"story","multilink",{"type":23,"attrs":169,"content":170},{"backgroundColor":25},[171],{"type":129,"content":172},[173,180,187,194],{"type":33,"content":174},[175],{"type":36,"attrs":176,"content":177},{"textAlign":25},[178],{"text":179,"type":41},"TCP/IP, VLANs, L2 and L3 switching",{"type":33,"content":181},[182],{"type":36,"attrs":183,"content":184},{"textAlign":25},[185],{"text":186,"type":41},"ECMP across parallel paths",{"type":33,"content":188},[189],{"type":36,"attrs":190,"content":191},{"textAlign":25},[192],{"text":193,"type":41},"Linux CLI: ip link, ethtool, jq",{"type":33,"content":195},[196],{"type":36,"attrs":197,"content":198},{"textAlign":25},[199],{"text":200,"type":41},"REST basics: curl, JSON, HTTP status codes",{"type":23,"content":202},[203],{"type":36},{"type":23,"content":205},[206],{"type":36},{"type":23,"attrs":208,"content":209},{"backgroundColor":25},[210,217,222,224,231],{"type":36,"attrs":211,"content":212},{"textAlign":25},[213],{"text":214,"type":41,"marks":215},"NETWORK AND INFRASTRUCTURE ENGINEER",[216],{"type":44},{"type":36,"attrs":218,"content":219},{"textAlign":25},[220],{"text":221,"type":41},"You have run TCP/IP, VLANs, and Ethernet switching for years. RDMA and lossless fabric are new, and the instinct that retransmission will cover a dropped packet is the assumption this course spends the most time correcting. You are comfortable at a Linux terminal and with a REST API.",{"type":36,"attrs":223},{"textAlign":25},{"type":36,"attrs":225,"content":226},{"textAlign":25},[227],{"text":228,"type":41,"marks":229},"CLUSTER OPERATIONS AND SUPPORT",[230],{"type":44},{"type":36,"attrs":232,"content":233},{"textAlign":25},[234],{"text":235,"type":41},"You are on the receiving end of \"the training job got slow.\" You need the investigation workflow: which telemetry decides, which only corroborates, and what the counter signature of a degraded cable looks like next to a PFC mismatch.",{"type":23,"attrs":237,"content":238},{"backgroundColor":25},[239],{"type":36,"attrs":240},{"textAlign":25},{"type":23,"attrs":242,"content":243},{"backgroundColor":25},[244],{"type":36,"attrs":245,"content":246},{"textAlign":25},[247],{"text":248,"type":41},"A hands-on course in the fabric that carries GPU-to-GPU traffic. RDMA does not tolerate packet loss the way TCP does, so the fabric has to be made lossless at the switch and the NIC, and it has to be kept that way. You validate RDMA host readiness, read and set lossless configuration with mlnx_qos, record a healthy bandwidth baseline, then diagnose two live fabric incidents from symptom to root cause without being told the fault class in advance, and fix one of them.","gpu-cluster-networking","training/courses/gpu-cluster-networking",-330,[],111882102,"c06c33d5-117c-408f-a685-ccc9b788bcbe","default","gpu-cluster-networking/",[],1789730290007]