AI Data Center Network Architecture Needs 800G/400G Transceivers

As AI technology and related applications advance, the significance of large models, big data, and AI computing power continues to grow in importance for the development of AI. Large models and datasets form the software foundation for AI research, while AI computing power is the most crucial infrastructure. In this article, we will explore how the development of AI affects data center network architecture. What does this mean for the 800G/400G transceivers market?

AI Drives the Upgrade of Data Center Network Architecture

The Fat Tree Data Center Network Architecture

As AI large model training becomes more widely used in numerous industries, traditional networks cannot match the bandwidth and latency requirements of large model cluster training. Large model distributed training necessitates communication between GPUs, which increases east-west traffic in AI/ML data centers, with a traffic pattern distinct from traditional cloud computing. AI data is short-term and high-volume, resulting in network latency and lower training performance in traditional network architecture. So, the emergence of the Fat Tree network is inevitable to meet the short-term and high-volume data processing demands.

In the traditional tree network topology, bandwidth converges layer by layer, with the network bandwidth at the tree's base being much less than the total bandwidth of all the leaves. In comparison, the fat tree looks like a real tree, with thicker branches closer to the root. As a result, network bandwidth increases from the leaf to the root, improving network efficiency and accelerating training. This is the underlying premise of the fat tree architecture, allowing for non-blocking networks.

Data Center Network Rate Upgrade Evolution

The demand for network speed grows in pace with the growing complexity of data center applications. From 1G, 10G, and 25G in the past to 100G, which is now extensively used, the rate of data center network upgrade and evolution is growing. However, in the face of large-scale AI workloads, 400G and 800G network rates have emerged as the next critical step in the evolution of data center networks.

The Rise of 800G/400G Transceivers Driven by AI Data Center

Reasons Behind the Growing Demand for 800G/400G Transceivers

Large-Scale Data Processing Demands

Training and inference of AI algorithms necessitate extensive datasets. Hence, data centers must efficiently handle the transmission of substantial data. The advent of 800G transceivers provides increased bandwidth, aiding in addressing this challenge. The upgraded data center network architecture typically includes two levels extending from the switch to the server, with 400G functioning as the bottom layer. So, upgrading to 800G will undoubtedly boost demand for 400G.

Real-Time Requirements

In certain AI application scenarios, the demand for real-time data processing is crucial. For example, in autonomous driving systems, the copious data generated by sensors necessitates prompt transmission and processing. Optimizing system latency becomes a pivotal factor in ensuring timely responses. The introduction of high-speed optical modules contributes to swiftly meeting these real-time demands by reducing the latency in data transmission and processing, thereby enhancing the system's responsiveness.

Multitasking Concurrency

Modern AI data centers often need to concurrently process multiple tasks, encompassing activities such as image recognition and natural language processing. The incorporation of high-speed 800G/400G optical transceivers enhances support for such multitasking workloads.

Bright Prospects for the 800G/400G Optical Module Market

The current demand for 400G and 800G optical modules has not experienced a significant surge; however, a notable upswing in 2024 is anticipated, primarily propelled by the escalating demand from AI computing. According to the authoritative market analysis firm, Dell'Oro, there is a forecasted increase in the demand for 400G optical modules in 2024. The heightened requirements for high-speed data transmission driven by AI, big data, and cloud computing are expected to catalyze the growth of the 800G optical module market. This trend highlights the bright future of the 800G/400G optical module market, indicating a path of greater adoption and usage in response to the changing demands of advanced computing applications.

Typical 800G/400G Transceiver Solution in Data Center

The picture illustrates the solution for upgrading to an 800G data center. The FS 400G optical transceiver QDD-FR4-400G and the 800G optical transceiver QDD800-2FR4-C1 form high-bandwidth links between the MSN4410-WS2FC switch, operating at a 400G interface speed in the backbone layer, and the high-performance 800G switch in the core layer.

Because these transceivers are packaged in high-density QSFP-DD, they can be deployed in high-density configurations. This increases transmission capacity and provides a greater bandwidth rate. Furthermore, by employing PAM4 modulation and retime technology, the transceivers achieve faster data transfer rates while considerably lowering latency. This upgrade improves overall system performance.

The Dawn of 800G/400G Transceiver is Approaching

As the need for faster and more efficient data transfer continues to rise, we find ourselves in the era of 800G/400G optical transceivers. These transceivers, known for their outstanding bandwidth capabilities, advancements in LPO technology, and cost-effectiveness, have the capacity to revolutionize the AI sector and redefine data centers. The utilization of high-speed optical transceivers brings us one step closer to unlocking the complete potential of AI.