Learning path

HPC networking from hardware to routed AI fabrics

Browse the full path in public. Sign in to open chapters and labs, sync progress, and participate in the technical discussion.

28

chapters mapped

14

labs ready to launch

Guest

public browsing mode

Part 1 - Foundations

No prerequisites - Ch 0-2

3 chapters

Part 2 - Fabric Operations

Requires Part 1 - Ch 3-8

6 chapters
3

The CLI - Reading the Fabric

The commands and discipline for reading HPC fabric state. Which commands run where, how to read their output, and the investigation workflow from physical layer to configuration.

45 minSign in required
ibstatshow dcb pfcethtoolDiagnostic workflow
Sign in to unlock->
4

InfiniBand Operations - ONYX CLI and Fabric Management

The InfiniBand operations layer: ONYX CLI, error counter interpretation, Subnet Manager management, ibdiagnet fabric sweep, and UFM event correlation.

50 minSign in required
ONYXibdiagnetUFMSubnet ManagerError counters
Sign in to unlock->
5

PFC, ECN, and Congestion Control

How losslessness actually works: PFC mechanics at the wire level, pause storm formation, ECN CE bit marking, DCQCN rate control algorithm, and the complete RoCEv2 port configuration checklist.

55 minSign in required
PFCECNDCQCNPause stormCongestion
Sign in to unlock->
6

Efficient Load Balancing

Why AI traffic is structurally low-entropy and how that breaks ECMP. The four load balancing modes (SLB, DLB, GLB, sDLB). Per-packet spraying and RSHP. In-cast congestion patterns and how to diagnose them from spine utilisation counters.

50 minSign in required
ECMPDLBGLBRSHPFlowletsIn-castElephant flows
Sign in to unlock->
7

Topology Design

How AI fabric scales from one switch to a SuperPOD. Fat-tree topology math, bisection bandwidth, BasePOD vs SuperPOD reference designs, oversubscription calculations, ROD vs RUD wiring, switch buffer selection, and cabling constraints.

60 minSign in required
Fat-treeBasePODSuperPODOversubscriptionRODRUDCabling
Sign in to unlock->
8

NCCL - The Application Layer

How NCCL translates AllReduce into RDMA operations. Ring vs Tree vs Double-Binary Tree algorithms. The environment variables that determine whether NCCL finds RDMA or falls back to TCP. Reading nccl-tests output. Correlating busbw degradation to fabric diagnostics.

55 minSign in required
NCCLAllReducebusbwnccl-testsDCQCN tuning
Sign in to unlock->

Part 3 - Physical Layer and Infrastructure

Requires Part 2 - Ch 9-11

4 chapters

Part 4 - Scale and Architecture

Requires Part 3 - Ch 12-16

9 chapters
12

Scale-Up Networking - NVLink Switch System

External NVLink Switch modules, 57.6 TB/s all-to-all at 256 GPUs, NVLink Network addressing, scale-up vs scale-out architecture decisions, and NVLink Switch diagnostics.

45 minSign in required
NVLink Switch57.6 TB/sScale-upNVLink NetworkSHARP
Sign in to unlock->
13

Alternative Topologies

Torus, folded torus, dragonfly, and TPU Pod design choices - where they came from, what workloads they suit, and why fat-tree remains dominant for AI training clusters.

45 minSign in required
TorusDragonflyGoogle TPUFat-tree
Sign in to unlock->
15

IP Routing for AI/ML Fabrics

How modern AI fabrics use routed Ethernet: BGP unnumbered, ASN design, BGP DPF, RIFT comparisons, Flex Algo, SRv6 path steering, and multi-tenant EVPN-VXLAN design.

55 minSign in required
BGPRIFTFlex AlgoSRv6Multi-tenancy
Sign in to unlock->
16

The GPU Compute Network - Packet Anatomy

A packet-level walkthrough from NCCL work queue entries to remote DMA completion: DGX interfaces, Queue Pair mechanics, ConnectX-7 processing, switch forwarding, and end-to-end packet decode.

55 minSign in required
Packet anatomyQueue PairsConnectX-7Leaf/spine forwardingWireshark
Sign in to unlock->
17

Storage Network Packet Path

A packet-level walkthrough of a checkpoint write from GPU HBM to storage appliance: NVMe-oF capsules, DMA paths, storage-fabric behavior, frame anatomy, and diagnostics.

55 minSign in required
NVMe-oFGDSPacket anatomyStorage fabricCheckpointing
Sign in to unlock->
18

OOB and Management Network

The out-of-band management fabric: BMC architecture, IPMI and Redfish internals, OOB topology, switch management isolation, UFM communication paths, BlueField-3 management architecture, and hardening guidance.

60 minSign in required
OOBBMCIPMIRedfishUFMManagement network
Sign in to unlock->
19

IP Addressing and Planning

A complete addressing reference for DGX BasePOD and SuperPOD deployments: address families, RFC 1918 partitioning, loopback design, P2P links, /32 server routes, management-plane planning, VXLAN VNI allocation, and scaling pitfalls.

65 minSign in required
IP addressingLoopbacksRFC1918BGP unnumberedVXLANSuperPOD
Sign in to unlock->
20

Ultra Ethernet Consortium (UEC)

Why RoCEv2 has friction at scale, and how UEC addresses it: UET packet format, SACK-based reliability without lossless fabric, 1-RTT congestion feedback, native multipath spraying, switch requirements, and honest deployment readiness as of March 2026.

60 minSign in required
UECUETPFC-freeSACK retransmitPacket sprayingNPM
Sign in to unlock->
21

Congestion Control Deep Dive

A rigorous, algorithm-level treatment of every congestion control scheme used in production AI fabrics — DCQCN, Swift, HPCC, TIMELY, and UEC CC — with practical guidance on parameter tuning and algorithm selection.

60 minSign in required
DCQCNSwiftHPCCTIMELYUEC CCCongestion ControlJCT
Sign in to unlock->

Part 5 - Advanced Networking

Requires Part 4 - Ch 22-23

2 chapters

Part 6 - Platform Integration

Requires Part 5 - Ch 24+

4 chapters

Labs

Scenario-based simulator work. Sign in to launch a lab and keep your troubleshooting progress attached to one account.

Need help choosing where to start?->
0

Fabric CLI orientation

A no-incident calibration lab. Learn how UFM, DGX OS, Linux netdev names, and Cumulus NVUE each describe the same two-rail fabric before you start troubleshooting failures.

12 minSign in required
CLIUFMDGX OSCumulusRail mapping
Sign in to unlock->
0

Read RoCE lossless counters

A healthy-link counter drill. Run a short RDMA write probe, then correlate switch-side PFC and RoCE counters with DGX ethtool counters so ECN marks, PFC pauses, and zero-drop behavior become familiar before incident labs.

15 minSign in required
PFCECNRoCE countersib_write_bw
Sign in to unlock->
0

Identify the failed rail

A GPU rail has gone dark. Use the topology map and CLI tools to identify which rail, confirm the RDMA state, and isolate whether the fault is on the NIC or switch side.

12 minSign in required
CLIPhysical diagnosticsRail state
Sign in to unlock->
1

Fix the PFC misconfiguration

A RoCEv2 workload is experiencing retransmissions. PFC is misconfigured. Diagnose and fix it.

10 minSign in required
PFCRoCEv2CLI
Sign in to unlock->
2

Diagnose fabric congestion

Throughput has dropped 40%. Investigate using interface counters and ECN configuration commands.

15 minSign in required
CongestionECNCounters
Sign in to unlock->
3

Diagnose uneven spine utilisation

AllReduce throughput has dropped with no drops visible. Diagnose why spine links are uneven and fix load balancing to restore full training throughput.

15 minSign in required
ECMPLoad balancingSpine counters
Sign in to unlock->
4

Evaluate topology proposals

Two vendors propose different switch configurations for a 64-node DGX H100 cluster. Calculate oversubscription ratios, identify which proposal meets requirements, and submit a recommendation before the purchase order is signed.

15 minSign in required
OversubscriptionCapacity planningTopology
Sign in to unlock->
5

Diagnose NCCL transport fallback

A 16-node cluster shows 3 GB/s busbw instead of expected performance. All hardware is healthy. Diagnose why NCCL is using socket transport and fix the environment variable misconfiguration.

20 minSign in required
NCCLTransport fallbackEnv vars
Sign in to unlock->
6

Triage a silent fabric degradation

Training is 12% slower - but no hard errors anywhere. Use UFM port counters, DCGM GPU metrics, and switch counters to correlate a rising pre-FEC BER across three monitoring layers and identify the marginal physical connector before it becomes a full link failure.

25 minSign in required
MonitoringDCGMUFMTelemetry
Sign in to unlock->
7

Uncover the hidden pause storm

A switch looks healthy at a glance, but the NIC reveals a severe pause storm. Check both ends, identify missing ECN, and restore rate control before continuous PFC pauses collapse throughput.

15 minSign in required
PFCPause stormECN
Sign in to unlock->
8

Fix the PFC priority mismatch

PFC is enabled, but on the wrong traffic class. Cross-check NIC drops, PFC priority output, and RoCE DSCP-to-priority mapping, then move PFC protection back to priority 3.

18 minSign in required
PFCPriority mappingRoCE
Sign in to unlock->
9

Recover the err-disabled rail

A rail has gone err-disabled after a physical fault. Recognise the NIC-side active-state trap, confirm the switch port failure, replace the optic, clear err-disable, and verify full recovery.

20 minSign in required
OpticsErr-disableRail recovery
Sign in to unlock->
10

ECMP hotspot: BGP bandwidth community

A reduced-capacity spine is still receiving equal ECMP traffic. Identify the missing BGP Link Bandwidth community and restore weighted ECMP before PFC storms spread.

22 minSign in required
BGPECMPWeighted paths
Sign in to unlock->
11

BGP suboptimal routing: spine ASN design

A link failure triggers a bad 3-hop path because the spines use different ASNs. Trace the suboptimal route, unify the spine ASN design, and verify clean failover behavior.

24 minSign in required
BGPFailoverASN design
Sign in to unlock->