DFT 1.6T Transceivers and DAC/ACC Cables for AI/HPC Data Center Interconnects

The rapid scaling of AI and LLM clusters has accelerated the demand for 1.6T interconnects. As data center traffic has shifted from being dominated by North-South to 80% East-West traffic, the massive East-West traffic generated by AI clusters places unprecedented pressure on bandwidth, latency, and energy efficiency. Ciena's report shows that AI-driven applications will further accelerate network traffic growth, with DCI network bandwidth expected to grow at least six-fold in the next five years. Moreover, in AI clusters, thousands of GPUs work together on model training, requiring the transmission of massive amounts of data, and any packet delay or loss will significantly extend task completion time, making high-performance interconnects even more critical. At this stage, 1.6T technology has moved beyond the theoretical roadmap to become a deployable reality with standardized technical architectures. On the InfiniBand side, NVIDIA launched the Quantum X800 switches at the GTC Developer Conference. As part of the NVIDIA X800 series, it is one of the world's first network platforms capable of delivering up to 800Gb/s end-to-end throughput. The platform includes the NVIDIA Quantum Q3400 switch with 144 800Gbps ports and the ConnectX-8 SuperNIC. Compared to the previous generation, its switching bandwidth capacity has increased by 5 times, and its network computing power has increased by 9 times to 14.4 Tflops, which is specially designed for high-performance computing and AI workloads that require extremely low latency and high throughput. alongside DGX B300 and GB300 systems is driving the rapid expansion and deployment of 1.6T InfiniBand XDR networking. As an essential configuration upgrade of GB300, the 1.6T optical module replaces the previous 800G one, which can enhance the data processing capability of the system, providing the essential fabric performance required for next-generation AI workloads such as deep learning and big data analysis. For Ethernet, the IEEE 802.3dj Working Group is expected to finalize the latest version of the 1.6TbE standard in 2026 to support the complete Ethernet rate system from 200G, 400G, 800G to 1.6T. DFT meets these evolving infrastructure needs by offering a comprehensive 1.6T product portfolio for both For InfiniBand and Ethernet environments, we offer high-speed transceivers and advanced copper cables to build and scale AI networks.

DFT 1.6T Optical Transceivers

DFT 1.6T optical transceivers deliver 1.6 Tbps throughput via 8-lane 212.5G PAM4 modulation, leveraging advanced DSPs or silicon photonics (SiPh) technology for superior signal integrity and power efficiency. Similar to the industry-leading 1.6T OSFP-XD DR8 optical module, it adopts PAM4 modulation technology which realizes efficient spectrum utilization by transmitting 2 bits of information per symbol, doubling the bandwidth compared with the traditional NRZ modulation scheme. And just like the 1.6T optical module launched by DFT with self-developed 200G silicon photonic chips, the silicon photonics technology applied gives it advantages such as small size, high integration and low power consumption, making it suitable for high-speed data transmission scenarios such as data centers, cloud computing and artificial intelligence. To address diverse cooling needs, the portfolio offers both OSFP-IHS (Closed Finned Top) and OSFP-RHS (Open Finned Top) solutions. (Flat Top) form factors, ensuring native compatibility with NVIDIA Quantum-X800 and B300/GB300 series platforms. As an industry-first 800Gb/s end-to-end high-performance network platform, NVIDIA Quantum-X800 is tailored for NVIDIA Blackwell architecture products, optimized for over trillion-parameter GPU computing and AI infrastructure, with single switch supporting 144 800GB/s ports and 9x network computing capability compared to the previous generation. The B300/GB300 series platforms feature a socket-style design for the B300 GPU, simplifying maintenance and upgrades while enhancing production flexibility. With reach options spanning 500m (DR8) to 2km (2FR4), these modules provide scalable interconnects for critical AI compute layers, from rack-to-rack to inter-data center fabrics, perfectly matching the high-bandwidth, low-latency demands of large-scale AI and HPC workloads.

DFT 1.6T Deployment Scenarios

Layered Deployment Logic

Intra-Rack & Adjacent Rack (<3m): For short-reach connections under 3 meters, DFT 1.6T OSFP Closed Finned Top passive DAC cables support direct connections on NVIDIA Quantum-X800 switches and can be used to build high-density fabrics for the NVIDIA SuperPOD with DGX B300 platform. For 1.6T liquid-cooled switch environments, DFT also offers 1.6T OSFP flat top active copper cables, optimized for high-performance deployments.

Intra-Data Center (500m): DFT 1.6T OSFP-IHS DR8 modules enable high-speed interconnection between NVIDIA Quantum-X800 switches and seamlessly interoperate with 800G OSFP-RHS DR8 modules to provide reliable connectivity between Quantum-X800 switches and DGX B300 systems. This validated connectivity supports the deployment of NVIDIA DGX SuperPOD™ with DGX B300 systems. All configurations have been tested in real-world deployment environments to ensure proven feasibility and reliability. In contrast, the 1.6T DR8 OSFP-RHS transceiver is intended for 1.6T liquid-cooled switches.

 Inter-Data Hall & Short-Reach DCI (Up to 2 km): For long-range links up to 2 km, DFT 1.6T OSFP 2FR4 transceivers provide scalable switch to switch optical connectivity for NVIDIA Quantum X800 deployments—this switch features 72*1.6T OSFP ports and 115Tbps total bandwidth—spanning adjacent data centers and short reach DCI (Data Center Interconnect) scenarios. The solution, which is compatible with InfiniBand NDR and Ethernet protocols, enables high bandwidth links for InfiniBand or Ethernet deployments. Adopting technologies like CWDM4 multi-wavelength optical multiplexing and optimized PCB design to reduce signal loss and crosstalk, it ensures stable data transmission while maintaining low power consumption, and uses duplex LC interfaces for simplified fiber management.

DFT Integrated 1.6T Interconnect Capability for AI Clusters

DFT demonstrates end-to-end 1.6T interconnect capability by combining the complementary strengths of copper and optical connectivity to build a high-performance AI fabric. This unified approach leverages the low latency and power efficiency of 1.6T DAC/ACC for critical intra-rack and adjacent-rack links helping simplify high-density cabling and support thermal management within compute pods. In parallel, high-bandwidth 1.6T DR8 and 1.6T 2FR4 transceivers extend the fabric across intra-data-center and short-reach DCI distances. With this complete 1.6T portfolio, DFT enables a balanced architecture that supports scalable GPU clusters while improving overall power efficiency and reducing total cost of ownership.

Advantages of DFT 1.6T Products

The DFT 1.6T portfolio is engineered to provide a future-proof foundation for AI innovation, delivering measurable advantages through rigorous testing and optimized design.

Verified Reliability: DFT 1.6T OSFP-IHS optical transceivers undergo 100% real-platform validation on NVIDIA QuantumX800 switches and ConnectX8 NICs. The 1.6T DR8 OSFP-IHS transceiver is also verified for interoperability with NVIDIA/Mellanox MMS4A00 optics to reduce deployment risk and troubleshooting effort. In addition, stringent BER verification and high SMSR help preserve signal integrity and improve data transmission accuracy. For example, in the BER test of optical modules, as shown in relevant data, the bit error rate is 7.7x10-14 at room temperature (25℃) and 2.38x10-7 at high temperature (70℃) after testing. BER is an important quality index of optical fiber digital communication systems, and strict verification of it is one of the key test items to ensure the normal operation of optical modules, which effectively guarantees the accuracy of data transmission.

Power Optimization: By integrating low-power Broadcom 3nm/5nm DSPs or silicon photonics technology in transceivers, DFT reduces the power required for 1.6T links. For shortreach links, DAC and ACC provide efficient copper interconnect options that avoid additional optical transceivers, helping lower overall fabric heat load and ease cooling pressure in highdensity AI clusters.

Future Scalability: The DFT 1.6T portfolio is engineered to adapt to evolving architectures like NVIDIA DGX B300, which is is positioned to accelerate the industry's transition to high-density 1.6T networking tailored for the AI reasoning era. Deploying these scalable solutions today minimizes the frequency and cost of future bandwidth upgrades and enables smoother expansion of IT infrastructure as architectures evolve.

Adequate supply: DFT ensures rapid fulfillment for 1.6T products with nearly 1,000 units back in stock across global warehouses. This supply chain reliability safeguards project timelines and shortens product delivery lead times. Shortening deployment cycles, improving the deployment efficiency to support the rapid expansion of large-scale AI clusters. For example, for super large Internet companies, adopting higher communication rate solutions like 1.6T optical modules can not only meet the demand for greater computing power, but also reduce the number of data center constructions, further improve computing power efficiency, lower operation and maintenance costs and energy consumption in the later period, which effectively supports the expansion of large-scale AI clusters.

Conclusion

As AI workloads move toward trillion-parameter models and increasingly large GPU clusters—with mainstream trillion-parameter model training already requiring super-10,000-card computing clusters as standard infrastructure—1.6T connectivity is rapidly becoming a core building block for next-generation fabrics. Industry data shows that NVIDIA's more powerful B100, expected to launch in Q2 2024, corresponds to server network cards upgraded to 800G and switch-side optical modules upgraded to 1.6T, with a GPU-to-1.6T optical module configuration ratio of 1:2.5. Moreover, the 1.6T optical module industry standard has been formulated, and the industrial ecosystem is basically mature, further solidifying its core position. The accelerating shift toward InfiniBand XDR and 1.6T Ethernet highlights a new phase where bandwidth, thermal efficiency, and deployment readiness must advance together to sustain AI. As a core driver for AI evolution, InfiniBand has taken a dominant position in the TOP500 list of global supercomputers, with 44.4% of the systems adopting this technology. It boasts excellent performance features such as outstanding data throughput, low latency, and high efficiency. Meanwhile, Ethernet, as the most mature and widely used network interconnection technology, is also evolving toward higher bandwidth represented by 1.6T to meet the needs of AI development. innovation.

DFT is helping drive this transition with a comprehensive 1.6T interconnect portfolio. With verified platform validation, power-optimized designs, and timely delivery assurance backed by global supply capabilities, DFT empowers organizations to accelerate the implementation of high-performance AI fabrics that maintain operational excellence and adapt to future technological iterations. Explore the complete DFT 1.6T InfiniBand and Ethernet product portfolio or consult with our technical experts to optimize your next-generation infrastructure deployment.