Spectrum-X Platform
Use the official platform page for current architecture and product-family context.
PLATFORM GUIDE · NVIDIA SPECTRUM-X
Spectrum-X Optical Interconnect Selection Guide
Spectrum-X is an NVIDIA AI Ethernet platform, so optic selection should begin with the exact validated platform configuration—not with a generic claim that an 800G or 400G module is “Spectrum-X compatible”.
Validated StackMatch the active NVIDIA solution-stack release
Endpoint IdentitySwitch + SuperNIC / NIC / DPU
Port EvidenceRate, form factor, rail and breakout mode
Media EvidenceOfficial cable / transceiver support and project validation
Official Platform Context
Start from the NVIDIA platform definition
NVIDIA describes Spectrum-X as an AI-optimized Ethernet platform coupling Spectrum switches with Spectrum-X SuperNICs and related software. Current NVIDIA documentation also lists supported NIC / SuperNIC families and directs deployments to the Spectrum-X Solution Stack for supported platform combinations.
Validated Solution Stack
Use the current validated configuration for supported combinations and version alignment.
Selection Workflow
Seven gates before proposing a Spectrum-X interconnect
01 / RELEASEIdentify the validated stack releaseRecord the NVIDIA Solution Stack version that applies to the deployment.
02 / COMPUTEIdentify the server / GPU platformCapture the exact compute system and networking endpoint context.
03 / ADAPTERIdentify SuperNIC / NIC / DPUUse the exact NVIDIA adapter family and model from the target deployment.
04 / SWITCHIdentify the Spectrum-X switchRecord exact switch family, model and target port.
05 / PORT MODEConfirm rate, rail and breakoutDo not infer the port mode from module rate alone.
06 / MEDIACheck official cable / optic supportUse NVIDIA platform evidence before proposing a specific cable or transceiver.
07 / VERSIONFreeze firmware and software contextKeep the validated hardware and software combination traceable.
Current Reference Architecture Snapshot
Lock the Spectrum-X RA, Network Operator and plane mode together
Current NVIDIA Network Operator 26.7 documentation maps Spectrum-X RA 2.3 to 26.7.x and documents single-plane plus dual- and quad-plane architectures. Treat that version alignment as part of the interconnect evidence.
Single-plane
- Current RA documentation includes BlueField-3 SuperNIC, ConnectX-7 NIC and ConnectX-8 SuperNIC contexts.
- Use the exact adapter and switch profile from the active validated configuration.
- Do not assume a cable or optic supported in one endpoint generation is valid for another.
Dual / quad-plane
- Current RA documentation includes ConnectX-8 SuperNIC dual- and quad-plane contexts.
- Network Operator 26.7 documents hardware plane load balancing (hwplb) and software plane load balancing (swplb).
- For multiplane platforms, current RA 2.3 guidance recommends hwplb when the validated profile supports it.
Version boundary: RA version, Network Operator release, NIC profile, firmware and plane mode form one evidence set. A change in one layer can require re-validation of the interconnect path.
Spectrum-X Version Lock
Five fields to freeze before approving the optical BOM
01 / RAReference Architecture versionRecord the active Spectrum-X RA and validated stack release.
02 / NICNIC / SuperNIC identityExact ConnectX / BlueField family, model and port.
03 / PLANEPlane modeSingle, dual or quad plane plus none / hwplb / swplb where applicable.
04 / SWITCHSwitch and port profileExact Spectrum-X switch, cage, rate, rail and breakout mode.
05 / MEDIAValidated interconnect pathOfficially supported cable / transceiver evidence plus project sample result.
Evidence Boundary
What QIANWU will not infer from a generic product label
Not sufficient evidence
- “800G OSFP” without the exact Spectrum-X switch / port
- “400G Ethernet” without the adapter and rail topology
- Matching connector shape without validated platform support
- One successful link-up used as universal compatibility proof
Required evidence path
- Current NVIDIA validated configuration
- Exact switch and SuperNIC / NIC / DPU identity
- Port rate, form factor and mode
- Official cable / transceiver support where specified
- Project-specific sample and interoperability validation
Brand boundary: NVIDIA and Spectrum-X are NVIDIA trademarks. QIANWU uses public NVIDIA documentation as platform evidence and does not treat a generic third-party optic as NVIDIA-validated without supporting evidence.
RFQ Inputs
What to send for a Spectrum-X optical RFQ
01Solution stack versionIf known, include the validated configuration release.
02Switch + adapterExact switch and SuperNIC / NIC / DPU models.
03Link definitionPort mode, rate, reach, connector and quantity.
04Validation targetSample, test window and required deployment date.
Have a Spectrum-X interconnect requirement?
Send the exact platform, switch, SuperNIC / NIC / DPU, port mode, link distance and quantity. We will keep vendor evidence and project assumptions separate.
Planning the adapter-side optical path?
Use the ConnectX guide to verify the exact adapter endpoint, peer switch and NVIDIA-supported interconnect evidence.