Networking NVIDIA UFM Cable Validation Tool
Networking NVIDIA UFM Cable Validation Tool
Breadcrumbs

Installation Notes

Supported Fabric Types


The UFM Cable Validation Tool is compatible with the following fabric types:

Fabric Type

Description

InfiniBand

Supported on MLNX-OS Switches with NDR and HDR port speeds

XDR

Supported on NVOS Q3400-RA Switches with XDR port speed

Ethernet

Supported on Cumulus Switches and Ethernet Hosts

NVLink

  • Supported on NVOS Switches and compute nodes of GB200/GB300 rack types: 36x2, 36x1, and 72x1.

  • Supported on internal NVLink links in the rack

  • Hosts with external links as Ethernet or IB 

  • The external InfiniBand links will report BER status, but neighboring information will need to be monitored from the neighboring switch side only.

Supported Devices

Type

Model

XDR Switches

  • Q3400_RA

NDR Switches

  • MQM9700

HDR Switches 

  • MQM8700

Ethernet Switches

  • SN5600

  • SN5610

  • SN2201 - support also for copper interfaces

NVLink Switches

  • N5200_LD

  • N5201_LD

  • N5110_LD

  • N5112_LD

  • N5113_LD

  • N5400_LD

  • N5500_LD

  • N5700_LD

Ethernet GPU Servers

  • SYS-421GE-TNHR2-LC-TW008 (H100)

  • PowerEdge XE9680

NVLink GPU Servers

  • GB200 NVL

  • DGX GB200 Compute Tray

  • Dell Server 9712a

  • DGX B300

Sonic Switches

  • SN3700

  • SN3700C

  • SN4600

  • SN4600C

  • SN4700

  • SN5400

  • SN5600

Supported Systems and Deployments

The UFM Cable Validation Tool supports the following system and rack deployments:

Deployment

Supported CVT scope

GB200

Supports GB200 NVLink rack configurations, including 36x1, 36x2, and 72x1 rack types. Includes internal NVLink validation and external Ethernet or InfiniBand connectivity.

GB300

Supports GB300 NVLink rack configurations, including supported rack-scale NVLink topology and external Ethernet or InfiniBand connectivity.

NVL576

Supports NVL576 solution deployments, including topology loading, device discovery, connectivity validation, and reporting.

Vera Rubin

Supports Vera Rubin deployments with shuffle-cable and shuffle-box topologies, including the associated unified topology and patch information.

Deployment support depends on the corresponding switch, server, operating-system, and topology-file requirements documented in this guide. Refer to Supported Devices, P2P File, and Pre-Requisites for Agents for detailed configuration requirements.


Devices not included in this list will be supported by CVT but may not be displayed correctly in the rack view.

Pre-Requisites for Agents

Device Type

Requirement

Details

All

Container Runtime

Docker installed and running on Server acting as collector and the Switches/Hosts being monitored.

Network Access

Port 8251 open for communication between the collector and agents(switches and hosts being monitored).

SSH Access

SSH access required for initial agent deployment on switches and hosts.

Servers

Architecture

x86_64 or aarch64

Packages

Servers should have mft  and ofed  packages installed

Ethernet Servers

Packages

Server should have lldpd installed and /var/run/lldpd.socket

GB200 compute Nodes

Packages

Should have nvidia-container-toolkit  and nvidia-smi  packages installed

SONiC Switches

LLDP Container

SONiC LLDP container running (docker ps | grep lldp)

SONiC Version

SONiC with config_db.json support


SONiC-Based Switches Support

The Cable Validation Tool (CVT) provides comprehensive support for SONiC (Software for Open Networking in the Cloud) based switches. CVT continuously monitors cable connections on SONiC switches, validates them against expected topology, and reports any discrepancies such as missing cables, wrong connections, or signal quality issues.

The CVT agent installer automatically detects SONiC switches by checking for the SONiC version file. No manual configuration of the agent type is required.


Last updated: