Our 800G DR8, SR8 and FR8 modules based on 8x100G PAM4 architecture are now in volume production, targeting AI training clusters and high-performance computing interconnect where port density and power efficiency matter most.
Why 800G, and why now
AI training clusters scale by adding accelerators, and every accelerator added puts pressure on the optical layer. An 800G port moves eight times the traffic of a 100G port while occupying one cage, which matters both for switch faceplate density and for the number of fibers a fabric consumes.
Available configurations
| Part number family | Optics | Reach | Typical use |
|---|---|---|---|
| OSFP-800G-DR8 | 1310nm single-mode | 500m | Row-level interconnect in AI pods |
| OSFP-800G-SR8 | 850nm multimode | 100m | In-rack and adjacent-rack links |
| OSFP-800G-FR8 | 1310nm single-mode, CWDM | 2km | Campus and metro extension |
All three variants use 8x100G PAM4 signalling, support forward error correction, and are offered in both DSP and LPO architectures so you can trade reach against power depending on the link budget.
Compatibility and ordering
- Backward compatible with lower-rate ports through breakout and fan-out assemblies.
- Matched AOC and MPO trunk options from the same supply chain, tested together with the optics.
- Sample units are available so you can validate performance on your own fabric before volume commitment.
Send us the switch platform and the link distances you are planning, and our engineering team will confirm the matching part numbers and quote accordingly.
Questions about this topic?
Send us the equipment model and the link requirements you are working with — our engineering team will recommend the matching configuration.
Request technical advice