COMPATIBILITY
ALL TRADEMARKS, REGISTERED COMPANIES & REFERENCES CITED ARE THE SOLE PROPERTY OF THEIR RESPECTIVE COMPANY AND ARE USED SOLELY TO ASSIST IN THE IDENTIFICATION OF PRODUCTS.
- NVIDIA MMS4A20-XM800(980-9IAT0-00XM00) COMPATIBLE
US$899.00
| Form Factor | OSFP224(Flat Top) | Data Rate | 850Gbps |
|---|---|---|---|
| Media | SMF | Wavelength | 1310nm |
| Connector | MTP/MPO-12 APC | Distance | 500m |
| Transmitter Type | EML | Receiver Type | PIN |
| Electrical Modulation | 4x 212.5G PAM4 | Optical Modulation | 4x 212.5G PAM4 |
| Power Consumption | ≤15.0W | Case Operating Temperature | 0 to 70°C (32 to 158°F) |
| Protocol | 800 Gigabit Ethernet 800G RoCE 800G Infiniband XDR | Application | NIC to Switch NIC to NIC Other Optical Links |
ALL TRADEMARKS, REGISTERED COMPANIES & REFERENCES CITED ARE THE SOLE PROPERTY OF THEIR RESPECTIVE COMPANY AND ARE USED SOLELY TO ASSIST IN THE IDENTIFICATION OF PRODUCTS.
Browse common switch series and models supported for each brand.
| Series | Compatible Switch Models (Examples) |
|---|---|
| ConnectX-8 | C8180 (900-9X81E-00EX-ST0)C8180 (900-9X81E-00EX-DT0)C8180L (900-9X81E-00EX-SL0)C8280Z (900-9X86E-00CX-ST0) |
800G DR4 OSFP224 Flat Top PAM4 1310nm 500m DOM MTP/MTP/MPO-12 APC InfiniBand XDR Optical Transceiver Module | 1.6T 2xDR4/DR8 OSFP224 Finned Top PAM4 1310nm 500m DOM Dual MTP/MPO-12 APC InfiniBand XDR Optical Transceiver Module | 1.6T 2xFR4/FR8 OSFP224 Finned Top PAM4 1310nm 2km DOM Dual LC Duplex InfiniBand XDR Optical Transceiver Module | |
|---|---|---|---|
| Form Factor | OSFP224(Flat Top) | OSFP224(Finned Top) | OSFP224(Finned Top) |
| Data Rate | 850Gbps | 1700Gbps | 1700Gbps |
| Connector | MTP/MPO-12 APC | Dual MTP/MPO-12 APC | Dual Duplex LC UPC |
| Distance | 500m | 500m | 2km |
| Electrical Modulation | 4x 212.5G PAM4 | 8x 212.5G PAM4 | 8x 212.5G PAM4 |
| Optical Modulation | 4x 212.5G PAM4 | Dual 4x 212.5G PAM4 | Dual 4x 212.5G PAM4 |
| Power Consumption | ≤15.0W | ≤25.0W | ≤26.0W |
Find answers to common questions about our products.
This 800G XDR transceiver is designed for AI data centers, high-performance computing clusters, large-scale GPU training, and inference workloads. It is suitable for network architectures that require high bandwidth, low latency, and high port density, especially for medium to large AI cluster upgrades and next-generation DGX B300/GB300 server interconnects.
Yes. This module is suitable for NVIDIA DGX B300 and DGX GB300 servers equipped with ConnectX-8 network adapters. It is typically deployed on the server-side 800G link and can work with 1.6T DR4-type transceivers on the switch side for high-performance server-to-switch connectivity in AI clusters.
An 800G DR4 module is typically a single-port 800G transceiver based on 4×200G PAM4 lanes and is mainly used on the server or NIC side. A 1.6T 2xDR4 module usually provides two 800G optical ports and is more commonly used on the switch side. In simple terms, 800G DR4 is mainly for NIC-side connectivity, while 1.6T 2xDR4 is mainly for switch-side connectivity.
Yes. With the correct speed, form factor, optical interface, and fiber connection, this module can interoperate with corresponding NVIDIA original DR4 or 2xDR4 transceivers. Before deployment, it is recommended to verify the switch model, NIC model, firmware version, module coding, and cable type to ensure stable link performance.
AICPLIGHT can support bulk delivery based on project requirements. Before shipment, the modules can be tested for compatibility, optical performance, aging, and link stability. For large AI cluster deployments, it is recommended to confirm the order quantity, target device models, deployment timeline, and testing requirements in advance to better support production planning and delivery.
We compared these with MMS4A20-XM800 optics in our setup, and the real-world performance has been very close. Stable links and no surprises so far.
Thermal performance has been solid in our dense racks. No overheating, no slowdown, and no unexpected alerts during operation.
We have been using them for several weeks now, and stability has been excellent. For AI cluster networking, this feels like a safe choice.