NVIDIA 800G OSFP 2xSR4 InfiniBand/Ethernet Transceiver

The NVIDIA 800G OSFP 2xSR4 InfiniBand/Ethernet Transceiver (MMA4Z00-NS-FLT) is an 800 Gigabit per second optical transceiver module designed for next-generation AI training infrastructure and high-performance computing fabrics. Built on an OSFP (Octal Small Form-factor Pluggable) form factor, the module leverages a 2xSR4 optical architecture — delivering two independent 400G SR4 channels over multimode fiber — enabling seamless integration into both InfiniBand NDR and high-speed Ethernet switching environments within the same physical interface.

The MMA4Z00-NS-FLT is optimized for short-reach data center interconnects using OM4 multimode fiber, supporting aggregated 800G bandwidth with low latency characteristics essential for tightly coupled GPU-to-GPU and GPU-to-switch communication patterns found in large-scale AI model training clusters. The transceiver supports breakout configurations, allowing network architects to split a single 800G port into 2x400G or further into lower-speed links, providing flexibility across spine-leaf and fat-tree topologies common in AI pod deployments.

Launched in early 2024, this transceiver targets hyperscale operators, cloud service providers, and enterprise HPC teams building or expanding AI training pods based on NVIDIA Quantum-2 or Spectrum-X switching platforms. The FLT suffix designates flat-top fiber optic technology, and the module is engineered to meet the power efficiency and port-density demands of modern AI data centers where thermal budgets and rack space are critical constraints.

Ideal for

• Large-scale GPU cluster interconnects for distributed AI model training, providing 800G lossless fabric bandwidth between compute nodes and top-of-rack switches
• InfiniBand NDR fabric deployments connecting NVIDIA DGX and HGX systems within AI supercomputing pods requiring deterministic, ultra-low-latency communication
• High-speed Ethernet spine-leaf architectures in AI data centers using NVIDIA Spectrum-X platforms where 800G port density is required at scale
• Breakout configurations enabling a single 800G OSFP port to serve multiple 400G or lower-speed downstream devices, optimizing switch port utilization in heterogeneous environments
• HPC research facility network upgrades where existing 400G infrastructure must be scaled to 800G to support growing parallel workload bandwidth demands
• Colocation and cloud provider backbone interconnects requiring short-reach, high-density optical links between adjacent racks or rows within the same data hall

Technical specifications

Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.