
Data Center Fiber Cabling for 400G & 800G
Select high-density MPO and LC assemblies from the exact transceiver MDI, fibre type, polarity, lane map, reach and channel-loss budget.
Technical page updated 2026-07-23
Quick answer
Data center fibre cabling must be selected from the exact transceiver optical interface—not from link speed alone. Confirm the MDI, fibre type, connector polish and gender, lane map, polarity, reach and complete channel-loss budget before selecting an MPO or LC assembly.
Suitable project types
- Leaf-spine, GPU-cluster and high-performance-computing interconnects
- Structured cabling for 400G and 800G Ethernet or InfiniBand designs
- MPO trunk, breakout and duplex LC channel planning
- Single-mode and multimode migration studies
Inputs required before selection
- 1Switch, adapter, DPU and transceiver manufacturer part numbers
- 2Ethernet or InfiniBand protocol and required breakout mode
- 3Optical MDI, lane count, connector polish, gender and key orientation
- 4OS2, OM4 or OM5 fibre requirement and total channel length
- 5Patch-panel count, mating points, pathway and cable-management limits
Planning workflow
- 1
Freeze the endpoints
Record the exact equipment ports and transceiver part numbers at both ends, including any split-mode restrictions.
- 2
Translate optics into fibre lanes
Map transmit and receive lanes to the required MPO or LC interfaces, fibre count, polish, gender and polarity.
- 3
Design the channel
Choose direct attach or structured cabling, then account for trunks, cassettes, patch cords, mating points, length and pathway density.
- 4
Validate before release
Check the manufacturer connectivity guide, loss budget, polarity drawing, labelling and acceptance test plan before ordering.
Selection checklist
- MPO-12, MPO-16 or duplex LC interface and actual populated fibre positions
- APC versus UPC polish and connector keying
- Male or female connector and transceiver pin requirement
- Straight or breakout lane map and end-to-end polarity
- Channel insertion loss, mating points, reach and cleaning access
Important limits
- A passive cable does not determine Ethernet or InfiniBand speed, reach or compatibility.
- MPO connectors with different polish types must not be mated.
- The same nominal speed can use different optical interfaces; every design requires the exact transceiver documentation.
Products mapped to this workflow
These products are relevant to specific stages of the workflow. Final suitability depends on the configuration and acceptance criteria above.

8-Fiber OM4 MPO-12/APC Type-B Cable for AI Clusters
Low-loss 8-fiber OM4 cable with MPO-12/APC female connectors and Type-B polarity for short-reach GPU, DPU, switch, and storage parallel-optics links.
View Details
8-Fiber OS2 MPO-12/APC Type-B Cable for 400G/800G DR Optics
8-fiber OS2 single-mode MPO-12/APC Type-B assembly for inter-rack and leaf-spine parallel-optics links using DR-class transceivers.
View Details
OM4 MPO-12/APC 1-to-2 Breakout Cable
OM4 multimode 1-to-2 MPO-12/APC breakout assembly for converting a four-channel parallel-optics path into two two-channel branches in supported OSFP and QSFP112 architectures.
View Details
OS2 MPO-12/APC 1-to-2 Breakout Cable
OS2 single-mode 1-to-2 MPO-12/APC breakout assembly for supported DR parallel-optics designs that split one four-channel path into two two-channel branches.
View Details
16-Fiber OM4/OM5 MPO-16 Cable for Native 800G Parallel Optics
16-fiber OM4 or OM5 MPO-16 cable assembly for native eight-lane multimode parallel-optics interfaces such as 800GBASE-VR8 and 800GBASE-SR8.
View Details
16-Fiber OS2 MPO-16 Cable for 800G DR8 and Emerging 1.6T
16-fiber OS2 MPO-16 parallel-optics assembly for 800G DR8-class links and infrastructure planning for emerging 1.6T optical interfaces.
View Details
OS2 Duplex LC Fiber Patch Cord for FR4/LR4 and DCI
Duplex OS2 LC/UPC patch cord for wavelength-multiplexed FR4, LR4, and data center interconnect links where the selected optical modules use duplex LC interfaces.
View DetailsRelated technical guides
Primary references
- IEEE 802.3df-2024 — 400 Gb/s and 800 Gb/s Ethernet operation
IEEE Standards Association
- Structured cabling requirements for InfiniBand and 400/800G Ethernet
NVIDIA Networking
Frequently asked questions
It depends on the transceiver. Designs may use multimode or single-mode fibre and may terminate in MPO or duplex LC interfaces. Use the exact transceiver MDI and reach specification.
No. Different polish geometries should not be mated. Confirm polish, keying, gender and lane map at both endpoints.
No. Fibre count alone does not establish compatibility. The optical engine, lane rate, MDI, polarity, loss budget and breakout design must all match.
Send the project inputs for configuration review
Include endpoint models, interfaces, quantities, route or process constraints, required test documents and acceptance criteria.
