Guide• NVIDIA

NVIDIA AI Networking Sourcing Guide

Commercial guide for buyers evaluating InfiniBand switches, Ethernet fabrics, adapters and interconnect planning for AI clusters.

Request support Open docs hub

Docs paths

Products

Type

Guide

Why buyers use this resource

A clearer reading order across NVIDIA documentation before commercial follow-up

Turns technical reading into purchasing steps and RFQ inputs

Reduces friction between official docs and sourcing action

Buyer interpretation

Use this resource to shorten review and improve RFQ quality

This page is written to interpret the technical topic in commercial terms, so the buyer can move faster from reading into model review, BOM preparation and quotation action.

Use this guide to align NVIDIA networking sourcing with your AI infrastructure roadmap. It helps commercial buyers structure part-number collection, compatibility review, lead-time communication and phased quotation requests for switching, adapters, optics and cables. It is especially useful when your project team needs a faster path from model research to RFQ submission.

What this resource helps validate

Key decisions this page is meant to support

The goal is not to replace official documentation, but to reduce uncertainty before your technical and purchasing teams send a structured inquiry.

Which NVIDIA product family best matches the project

What to review next inside official documentation

How to turn that into a more accurate inquiry

RFQ handoff

What to send after reading this page

A stronger inquiry usually includes the deployment target, preferred models, quantity and the exact compatibility or supply question that still needs confirmation.

The product family or models that seem closest to your need

Quantity range and a short deployment scenario

Any compatibility or delivery blocker that needs a quick answer

Start inquiry Browse products

Official docs references

Relevant NVIDIA documentation paths for deeper technical review

Use these official documentation areas as the technical source of truth, while keeping this page focused on buyer-side interpretation and RFQ preparation.

Drivers / firmware / tools

Networking Software

Manuals of adapter drivers, firmware, accelerators, switch operating systems, management, and tools.

Official docs

ConnectX adapters

Networking Adapters

Documentation for NVIDIA® ConnectX® family of networking adapters.

Official docs

DPUs / SuperNICs

BlueField DPUs / SuperNICs & DOCA

Documentation for NVIDIA's data processing units (DPUs) and SuperNICs.

Official docs

Cables / transceivers

Networking Interconnect

Documentation for NVIDIA® LinkX® interconnect family of cables and transceivers.

Official docs

Docs hub

Recommended products to continue the conversation

Continue with exact NVIDIA models

These products are practical follow-up points when the buyer wants to move from reading into exact part-number or BOM discussion.

MQM9700-NS2R

NVIDIA Quantum-2 QM9700-NS2R 64-Port 400G InfiniBand Managed Switch

64-port 400G InfiniBand managed switch with 51.2 Tb/s throughput, SHARPv3 acceleration, and ultra-low latency for AI/HPC clusters.

MQM9700-NS2R

Product photoCMS image ready

Ordering code: MQM9700-NS2R(920-9B210-00RN-0M2)
51.2 Tb/s aggregate bidirectional throughput; >66.5 billion packets per second (BPPS): x86 Coffee Lake i3, 8GB DDR4 SO-DIMM (2666 MT/s), 16GB M.2 SSD

View product Request quote

MSN3420-CB2F

NVIDIA Spectrum SN3420 MSN3420-CB2FC – 48-Port 25GbE & 12-Port 100GbE Top-of-Rack Switch with Cumulus Linux

High-performance 1U rack switch with 48x25GbE ports & 12x100GbE uplinks. Features 4.8Tb/s capacity, Cumulus Linux OS, and 425ns latency for data centers.

MSN3420-CB2F

Product photoCMS image ready

48x SFP28 (10/25GbE) + 12x QSFP28 (40/100GbE): 48 + up to 48 via breakout (total 96 logical 25GbE from QSFP28)
1x 100/1000 RJ45, 1x RJ45 serial, 1x USB: 2x hot-swappable (1+1 redundant), AC 100-264V, 50-60Hz

View product Request quote

MCX75310AAS-NEAT

NVIDIA ConnectX-7 MCX75310AAS-NEAT Dual-Port 400Gb/s InfiniBand & Ethernet Smart Adapter – PCIe 5.0 x16, NDR

High-performance 400Gb/s dual-port adapter with PCIe 5.0 x16, hardware security offloads, and NVMe-oF support for AI/HPC data centers.

MCX75310AAS-NEAT

Product photoCMS image ready

Ordering code: MCX75310AAS-NEAT(900-9X766-003N-SQ0)
PCIe 5.0 x16 (32 lanes): IPsec, TLS 1.3, MACsec, AES-XTS

View product Request quote

FAQ

Questions buyers may still ask

Short answers that help turn reading into a clearer sourcing conversation.

Can NVIDIA AI Networking Sourcing Guide be used before sending an RFQ?01

Yes. This resource is written to help buyers prepare clearer technical and commercial context before quotation starts.

Is official documentation alone enough for sourcing?02

Official docs are important for technical truth, but buyers also need purchasing interpretation that connects the content to part numbers, quantity and compatibility.

Can I move from this resource directly into product review or inquiry?03

Yes. After reading, you can continue to the relevant product page or send an inquiry with target models, quantity and project notes.

Related resources

Continue with adjacent buyer resources

Use the next resource to compare options, verify compatibility or narrow the shortlist before sending an RFQ.

ComparisonNVIDIA

QM9700 vs SN4700 Buying Comparison

A commercial comparison page built for buyers deciding between InfiniBand and Ethernet switching paths in AI and data center projects.

Read resource

CompatibilityNVIDIA

ConnectX Adapter Compatibility Checklist

Checklist for buyers reviewing NVIDIA ConnectX adapter generation, host interface, speed, cable type and driver context before quotation.

Read resource

CompatibilityNVIDIA

NVIDIA Optics and Cables RFQ Checklist

Practical checklist for preparing optics, DAC, AOC, breakout and cable-length details before asking for NVIDIA interconnect pricing.

Read resource

Next step

Ready to move from research into RFQ action?

Send your target models, quantity range, project scope or compatibility blocker and we can continue with quotation and sourcing support.

Request support