netcore
NVIDIA/Mellanox MMS4A20 Compatible 800GBASE-DR4 OSFP224 RHS/Flat Top 4 x 200G PAM4 1310nm 500m MPO-12/APC SMF InfiniBand XDR Transceiver Module
1 / 7
Optical Transceivers · OSFP224-DR4-800G-FL

NVIDIA/Mellanox MMS4A20 Compatible 800GBASE-DR4 OSFP224 RHS/Flat Top 4 x 200G PAM4 1310nm 500m MPO-12/APC SMF InfiniBand XDR Transceiver Module

NVIDIA/Mellanox MMS4A20 Compatible 800GBASE-DR4 OSFP224 RHS/Flat Top 4 x 200G PAM4 1310nm 500m MPO-12/APC SMF InfiniBand XDR Transceiver Module, Product Specification:NVIDIA/Mellanox InfiniBand Compatible - MMS4A20, Vendor Name - FS, Form Factor - Single-port OSFP RHS/Flat Top, Max. Data Rate - 850Gbps (4x 212.5Gbps), Wavelength - 1310nm, Max. Cable Distance - 500m

Net price · DE
€948,00
ex. VAT
In stock

NVIDIA/Mellanox MMS4A20 Compatible 800G OSFP224 DR4 8x 200G PAM4 InfiniBand XDR Optical Transceiver Module (SMF, 1310nm, 500m, MPO-12/APC, DOM, RHS/Flat Top)

The OSFP224-DR4-800G-FL is compatible with NVIDIA/Mellanox MMS4A20 and is designed for next-generation InfiniBand XDR and high-speed data centre interconnect. The 800G DR4 transceiver operates at 200Gb/s per lane with 4x 200Gb/s PAM4 modulation, supporting transmission distances of up to 500 metres, making it ideal for short-reach intra–data centre deployments. It complies with OSFP MSA, CMIS Rev 5.2, IEEE 802.3dj, and OIF 224G-VSR standards. The built-in digital diagnostics monitoring (DDM) allows access to real-time operating parametres.
Featuring an advanced Broadcom 3nm DSP and silicon photonics technology, the module delivers ultra-low BER, strong signal integrity, and high link reliability at 800G speeds while improving integration density and energy efficiency. It is used to link the Quantum-X800 QM3x00 switches with twin-port OSFP 2x800Gb/s transceivers to dual 800Gb/s ConnectX-8 mezzanine cards in DGX B300/GB300 liquid-cooled AI systems or ConnectX-8 PCIe cards.
Note: "Finned-top" transceivers are now called "Integrated HeatSink" (IHS) as the cooling fins are integrated into the OSFP connector metal top. Commonly used in air-cooled switches. "Flat-top" transceivers are now called "Riding Heat Sink" (RHS) as the cooling fins ride on top of the network card or systems internal cages. This is used for both air-cooled and liquid-cooled applications for DGX, ConnectX-7, and BlueField-3 DPUS.

Highlights

Specifications

26 fields
NVIDIA/Mellanox InfiniBand CompatibleMMS4A20
Vendor NameFS
Form FactorSingle-port OSFP RHS/Flat Top
Max. Data Rate850Gbps (4x 212.5Gbps)
Wavelength1310nm
Max. Cable Distance500m
ConnectorMPO-12/APC
MediaSMF
Transmitter TypeCW (SiPh-Based)
Receiver TypePIN
TX Power-3.3~4dBm
Minimum Receiver Power-6.3dBm
Power Budget3dB
Receiver Overload4dBm
Max. Power Consumption16W
Extinction Ratio3.5dB
DDM/DOMSupported
Commercial Temperature Range0 to 70°C (32 to 158°F)
Modulation (Electrical)4x 200G-PAM4
Modulation (Optical)4x 200G-PAM4
CDR (Clock and Data Recovery)TX & RX Built-in DSP
Inbuilt FECYes
Packaging TechnologyCOB (Chip on Board) Packaging
Bit Error Ratio (BER)1E-12(FEC Enabled)
ProtocolsOSFP MSA, CMIS Rev 5.2, IEEE 802.3dj, OIF 224G-VSR
Warranty5 Years

Features

Boosting B300 AI Cluster Performance with 800G OSFP224 Modules

  • In the NVIDIA B300 architecture, the 800G OSFP224 optical module provides high-bandwidth, low-latency connectivity between Quantum-X800 InfiniBand switches and DGX B300 servers, ensuring stable, high-speed interconnections, accelerating large-scale distributed AI training, and enhancing overall cluster efficiency. Learn more about the <a href="/de-en/solutions/1.6t-optical-interconnect-for-infiniband-b300-ai-clusters-s0101.html" >B300 InfiniBand solution</a>.

Broadcom 3nm DSP & SiPh-Based Technology for Superior Energy Efficiency

Powered by Broadcom's 3nm DSP and silicon photonics chip, this 800G XDR transceiver delivers low power consumption of 16W (over 11% power savings) and a tight 1E-12 BER (with FEC Enabled). It enables higher port density, reduced cooling requirements, and lower total cost of ownership, while seamlessly pairing with DGX GB300/B300 to scale smoothly for AI clusters.

XDR vs. NDR: Ultra-Large AI Cluster Efficiency Breakthrough

FS's XDR solution featuring the 800G OSFP224 transceiver boosts cluster energy efficiency through high bandwidth density and an optimised link architecture. In 512-node configurations, this solution enables unmatched throughput, advanced scalability, significantly reduced power and space requirements and a 40.52% cut in total costs.

Comprehensive Testing Increases Reliability

Qualified through a rigorous process with advanced equipments to ensure you get the high quality and reliable optics.

Reliable Test Result

  • Multi-Version
  • Connectivity
  • BER