Broadcom Sian3 | BCM83628 Broadcom Sian3 | BCM83628 Item Spotlights
NVIDIA/Mellanox MMS4A00-XM (980-9IAH1-00XM00) Compatible 1.6T 2xDR4/DR8 InfiniBand XDR Silicon Photonics Optical Transceiver Module (Twin-port OSFP224, Broadcom 3nm DSP | Sian3 BCM83628, 1310nm, 500m (OS2) , Dual MPO-12/APC, DOM, SMF, IHS/Closed Finned Top)
NADDOD OSFP-1.6T-2xDR4H optical transceiver is fully compatible with NVIDIA/Mellanox MMS4A00-XM (980-9IAH1-00XM00) and is purpose-built for next-generation InfiniBand XDR and high-speed Ethernet data center interconnect. This 1.6Tb/s dual-port OSFP224 transceiver adopts a silicon photonics–based parallel single-mode optical architecture, supporting 2 × 800G DR4 transmission over eight parallel single-mode fibers. Each 800G port operates with 4 × 200G-PAM4 lanes, delivering reliable high-speed data transmission with a maximum reach of up to 500 meters, ideal for short-reach intra–data center deployments. The 1.6T OSFP 2xDR4 module integrates two independent 4-lane optical engines, each terminated with an MPO-12/APC connector, providing a total of eight parallel optical lanes. This architecture is optimized for high port density, low latency, and power-efficient switching environments, where scalability and signal integrity are critical.
Leveraging advanced silicon photonics chip, the 1.6T OSFP 2xDR4 XDR optical transceiver delivers high integration density, improved power efficiency, and enhanced front-panel scalability. Powered by the Broadcom 3nm Sian3 DSP (BCM83628), the module features advanced PAM4 signal conditioning and link robustness with ultra-low bit error rates (BER), enabling stable, low-latency 800G and 1.6T optical interconnects. The NADDOD OSFP-1.6T-2xDR4H is specifically designed for connectivity between NVIDIA Quantum-X800 air-cooled switches, forming the optical foundation of InfiniBand XDR AI and HPC data center fabrics.
NADDOD 1.6T transceiver has been successfully deployed in real XDR AI clusters, maintaining continuous, stable operation under full-port, full-load conditions. Tested with the NVIDIA Quantum-X800 (Q3400-RA) switch and ConnectX-8 network card in typical high-bandwidth, continuous XDR workloads, this transceiver achieves an ultra-low BER of 5E-14 at a mere 25W power consumption. This fully validates its reliability in high-density, long-term full-load XDR cluster deployments.
As XDR networks scale in port density and bandwidth, power consumption, cooling performance, and link stability become critical challenges. NADDOD 1.6T transceiver features a collaborative design that combines Broadcom 3nm DSP (BCM83628) with in-house silicon photonics, and is already operating stably in real XDR clusters. Compared with common 5nm DSP solutions, this design delivers lower power consumption, more stable links, and better heat dissipation in high-density scenarios—enabling long-term full-load operation of large-scale GPU clusters, while cutting port failure rates and reducing operational costs.
In large-scale AI clusters, XDR networking based on this 1.6T transceiver delivers measurable gains over NDR architectures. Real operational data from a 512-node cluster shows a reduction of ~ 7 racks (20 kW × 42U), a total power reduction of 130.4 kW (≈ 16–26 GPUs’ power), and a total cost reduction of 41%. By reducing rack count, simplifying cabling, and improving space utilization, this solution increases compute density while enabling flexible future cluster expansion.
NADDOD 1.6T transceiver has been delivered at scale and validated on HGX B300 servers, fully compatible with NVIDIA DGX B300, GB300, and OEM HGX B200 high-density GPU servers. Supported by a mature production system and stable chip supply, it serves as a core solution for high-speed interconnect between servers and switches under the XDR architecture, and is already operating reliably in large-scale AI clusters, enabling smooth transition from validation to large-scale deployment.
| OSFP-1.6T-2xDR4H |  OSFP-1.6T-2DR4FH |  OSFP-1.6T-2xFR4H |  OSFP-1.6T-2FR4FH | |
|---|---|---|---|---|
| NVIDIA PN | MMS4A00-XM | MMS4A00-XM-FLT | MMS4A50-XM | / |
| Form Factor | OSFP224 | OSFP224 | OSFP224 | OSFP224 |
| Thermal Design | IHS/Closed Finned Top | RHS/Flat Top | IHS/Closed Finned Top | RHS/Flat Top |
| Media | SMF | SMF | SMF | SMF |
| Center Wavelength | 1310nm | 1310nm | 1310nm | 1310nm |
| Connector | Dual MPO-12/APC | Dual MPO-12/APC | Dual LC Duplex | Dual LC Duplex |
| Transmission Distance | 500m | 500m | 2km | 2km |
| Electrical Modulation | 8x212Gb/s PAM4 | 8x212Gb/s PAM4 | 8x212Gb/s PAM4 | 8x212Gb/s PAM4 |
| Optical Modulation | Two ports 4x212Gb/s PAM4 | Two ports 4x212Gb/s PAM4 | Two ports 4x212Gb/s PAM4 | Two ports 4x212Gb/s PAM4 |
| Power Consumption | ≤25W | ≤25W | ≤26W | ≤26W |
| Transmitter Type | DFB (SiPh Based) | DFB (SiPh Based) | DFB (SiPh Based) | DFB (SiPh Based) |
| Matching Cables | MTP-12 APC Type B OS2: S2MTPA12FB | MTP-12 APC Type B OS2: S2MTPA12FB | LC to LC: S2LCUD | LC to LC: S2LCUD |
| Application | 1.6T Air-Cooled Switch-to-Switch Interconnect; 1.6T Air-Cooled Switch–to–800G ConnectX-8 HCA Interconnect | 1.6T Liquid-Cooled Switch-to-Switch Interconnect | 1.6T Air-Cooled Switch-to-Switch Interconnect | 1.6T Liquid-Cooled Switch-to-Switch Interconnect |
5.0
5.0
5.0
5.0
5.0