AI Drives Surge In Demand For Intelligent Computing

At present, a new round of scientific and technological revolution and industrial transformation is accelerating, and digital technologies such as artificial intelligence, cloud computing, big data, and quantum information are developing rapidly. In particular, the emergence of large models represented by Chat GPT has promoted digital technology to accelerate into the era of artificial intelligence.

Data centers have become global infrastructure, with their number and scale constantly expanding, and the demand for data transmission and interconnection technology is also increasing. As a high-speed, reliable, and low-latency communication technology, optical interconnection technology has broad application prospects in the field of data centers.

The emergence of AI big models has driven the continuous development of new computing infrastructure construction, and traditional data centers have accelerated their transformation into intelligent computing centers. Data shows that by the end of 2023, China's intelligent computing power will exceed 230EFLOPS, surpassing general computing power in all aspects, and the gap is gradually widening.

The breakthrough progress of large artificial intelligence models represented by GPT has stimulated a global boom in the development of intelligent computing power. Clusters built with Qianka computing chips will become the standard for training large models with hundreds of billions of parameters. How to accelerate the construction of high-performance networks to support the development of intelligent computing power has become a hot topic in the industry.

The global demand for intelligent computing power is growing rapidly, and there is an urgent need to build a new intelligent computing center (NICC) technology system with unified standards, leading technology, software and hardware collaboration, and open compatibility. Among them, the intelligent computing center network is used to connect pooled heterogeneous computing resources such as CPU, GPU, and memory, and runs through the entire process of data computing and storage. The enhancement of network performance is of key significance to improving the overall computing power level of the intelligent computing center.

The emergence of AI big models has driven the continuous development of new computing infrastructure construction, accelerated the transformation of traditional data centers into intelligent computing centers, and put forward new development needs and challenges in terms of network scale, latency, bandwidth and reliability.

As the main participants in the large model business, Internet companies have their own understanding of the requirements for intelligent computing. Chen Qin, an optical network technology expert at Alibaba Group, said that AI large model training has extremely high requirements for the stability of optical interconnection, and the software and hardware of optical modules need to be iteratively upgraded. Baidu senior engineer Zhou Qian believes that the parameters of large models are growing very fast, and how to keep up with the pace of optical bandwidth in data centers has become a very important issue. In addition, the requirements for transmission stability between data centers are also higher. Chen Cheng, an optical interconnection architect at JD.com, said that the requirements of intelligent computing networks for optical interconnection are mainly concentrated on large bandwidth, low cost and low latency.

Facing the new demands and challenges that intelligent computing centers have put forward for the network. As a high-speed, reliable, and low-latency communication technology, optical interconnection technology has broad application prospects in the field of intelligent computing centers. At present, the industry's intelligent computing center high-performance network innovation is mainly divided into two directions. One is based on the optimization of existing Ethernet, and the other is to innovate the underlying Ethernet solution. The fully dispatched Ethernet (GSE) led by Chinese companies is an innovative solution that can maximize compatibility with the Ethernet ecosystem, build a new type of intelligent computing center network with non-blocking, high bandwidth, and low latency, and form a standard and open technology system to help the development of the AI industry.

FCST - Better FTTx, Better Life.

We have been designing, manufacturing and supplying passive components for fiber optical networks since 2003. Everything we do in FCST is designed to provide efficient, simple and innovative solutions to solve complicated tasks.

We offer 4 solutions

lFTTH ODN Solution

lUnderground Cable Installation Solution

lAerial Cable Installation Solution

lFiber Tools & Test Solution

As a FTTx solution supplier, we can also do ODM according to project. We will continue to innovate and contribute more to the construction of the global optical fiber network.