Specifications
Vendor
FS
Scale-out rates
400G, 800G and 1.6T transceivers
Scale-out fabrics
Ethernet, RoCE and InfiniBand between GPU servers, NICs, switches and racks
1.6T modules
OSFP-DR8-1.6T and OSFP-2FR4-1.6T twin-port OSFP
1.6T electrical interface
8x 200G PAM4 per module
1.6T DSP
Broadcom 3 nm DSP
DR8 optics
Silicon photonics with CW laser, dual 4x 200G at 1310 nm, dual MTP/MPO-12, 500 m, 25 W
2FR4 optics
EML transmitters on CWDM4 wavelengths, dual duplex LC, 2 km, 26 W
NVIDIA validation
100% verified on NVIDIA Quantum-X800 switches
NVIDIA part equivalents
MMS4A00 and MMS4A50-XM
Target link types
1.6T switch-to-switch and 1.6T-to-two-800G breakout into DGX B300 systems
Coherent / scale-across
800G ZR/ZR+ in OSFP and QSFP-DD, Marvell 5 nm DSP
Coherent reach
120 km and 500 km over single-mode fiber depending on line
Coherent compliance
800ZR, OpenZR+, OIF with Open FEC and CMIS/C-CMIS
Wavelength bands
C-band and L-band operation
D7070 muxponder
1U, 100GbE/400GbE client on QSFP28 and QSFP112 to four 800G coherent wavelengths via CFP2-DCO, 3.2 Tb/s
BOX 5 Ultra toolkit
100M to 1.6T across 15 form factors and 200+ brands
BOX 5 Ultra battery
3,200 mAh, rated up to 8 hours
OSFP top variants
Closed-finned and open-finned at ~13 mm, flat top at ~9.5 mm
Overview
FS has packaged its AI interconnect optics into a deliberately simple two-tier structure that maps to the two problems every GPU cluster build faces. Scale-out covers the links inside the cluster - between GPU servers, NICs, switches and racks across Ethernet, RoCE and InfiniBand - and tops out at 1.6T modules validated against NVIDIA Quantum-X800 switches. Scale-across covers the links between clusters, where coherent optics carry traffic over up to 500 km of single-mode fiber.
At the top of the scale-out stack sit the OSFP-DR8-1.6T and OSFP-2FR4-1.6T twin-port modules, each running 8x 200G PAM4 electrically on a Broadcom 3 nm DSP. The DR8 uses silicon photonics with a CW laser and dual MTP/MPO-12 connectors for 500 m reaches at 25 W; the 2FR4 uses EML transmitters on CWDM4 wavelengths for 2 km at 26 W. FS reports 100% verification on NVIDIA Quantum-X800 switches and lists them as equivalents to NVIDIA part numbers MMS4A00 and MMS4A50-XM, with the DR8 aimed at switch-to-switch links and 1.6T-to-two-800G breakouts into DGX B300 systems.
For inter-site capacity the 800G ZR/ZR+ coherent modules ship in OSFP and QSFP-DD with a Marvell 5 nm DSP, rated for 120 km and 500 km lines and supporting both C-band and L-band to raise capacity per fiber pair. Sites aggregating client traffic first can use the D7070 muxponder, a 1U platform taking 100GbE and 400GbE client services on QSFP28 and QSFP112 and placing them onto four 800G coherent wavelengths through CFP2-DCO pluggables for 3.2 Tb/s per platform. The BOX 5 Ultra handheld toolkit covers field provisioning - configuring, validating and wavelength-tuning modules from 100M to 1.6T across 15 form factors and 200+ brands, on a 3,200 mAh battery rated for eight hours.
Key Benefits
One optics source for both tiers covers intra-cluster scale-out and inter-site scale-across without a second vendor qualification. NVIDIA Quantum-X800 verification and MMS4A00/MMS4A50-XM equivalence removes compatibility risk on InfiniBand XDR fabrics. Three OSFP top profiles match the cooling style of each switch cage or NIC, including liquid-cooled gear. BOX 5 Ultra field provisioning means optics can be coded and validated on a bench rather than consuming production switch ports and change windows.
Applications
AI and GPU cluster fabrics, InfiniBand XDR scale-out, DGX B300 800G breakouts, data center interconnect and scale-across up to 500 km, cloud and carrier DCI aggregation, and on-site optical provisioning and validation.
Request a Quote — FS AI OPTICS PORTFOLIO — 400G TO 1.6T SCALE-OUT AND COHERENT MODULES
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →