Edge, a concept relatively far from the "center". Edge computing is often deployed outside the data center, close to the computing provided by the on-site application side, and the market it faces is more dispersed. Industry, retail, and smart cities are all typical scenarios in edge computing. A lot of data is too late or cannot be uploaded to the cloud in a centralized manner, so calculation and reasoning need to be completed on the edge terminal side.
The day before Intel (INTC.US) released its Q2 financial report, the 2023 Intel Network and Edge Industry Forum discussed the opportunities for future edge computing. Sachin Katti, senior vice president of Intel Corporation and general manager of the network and edge division, mentioned that the global edge market will reach a scale of 445 billion US dollars in 2030, and the promotion of enterprise demand will be transformed into the momentum of economic development. Therefore, he believes that the edge Development will become a huge development opportunity in the next 10 years . " The edge is already where the largest deployment of artificial intelligence is now ."
Sachin Katti
The hottest topic surrounding artificial intelligence in the past six months is large-scale models, and this meeting is no exception. The consensus is that the implementation of large-scale models should be combined with the industry, and targeted to solve industry problems.
The application of large models has different implementation methods: the training can be completed in the cloud, and then the inference can be completed in the cloud and the result returned to the edge side of the front end, but this means that the communication time may be longer; the training can also be completed in the cloud, and then in the After the fine-tuning is completed on the edge side, the inference is made and the result is given.
"In our experience, more than half of AI deployments fail." Sachin Katti believes that there are several reasons for this. The infrastructure problem is too old, and it is difficult to deploy large models of artificial intelligence to old equipment; The second is data. We need to learn how to capture isolated data and use these data for deployment. He said bluntly, "This is a very big challenge, and many people have stopped because of these challenges."
"For customers, it is very difficult to obtain marginal value through hardware alone. So when we deliver the hardware to customers, we will also deliver the required software to customers. Our purpose is to hope that through this A set of infrastructure can make it easier for customers to deploy our hardware and manage the hardware." Sachin Katti said.
He mentioned that Intel continues to make efforts in three directions in order to seize the opportunity of edge computing: In terms of hardware, Intel has just released the accelerator Gaudi 2 for large-scale model training in China, as well as other general-purpose CPU processors and Dedicated discrete graphics accelerator; in terms of software, OpenVINO can optimize and compress the large model after cloud training. In the latest version of OpenVINO, support for large model Stable Diffusion has been added; in addition, it is to continue to develop Ecological partners.
"The time cost and labor cost of deploying artificial intelligence to the factory production line are still high. If there is a more innovative idea on the large model, we will cooperate with the technical team and the product team to refine it together. We hope to implement it more quickly. Bring in Intel’s technical solutions.” Wang Zhicong, vice president of Intel’s marketing group and general manager of China, believes that business partners should capture the new momentum emerging in the recovery process of the Chinese economy, the rise of new industries, and the introduction of technology into the next development stage.
However, the road to opportunity is not smooth.
Intel's Q2 financial report for 2023 shows that the revenue in the second quarter was US$12.948 billion, a decrease of 15% compared with US$15.321 billion in the same period last year; $454 million. From the perspective of segmented fields, although the network and edge business has maintained a steady year-on-year growth in the past year, this business has declined for two consecutive quarters this year. The revenue for this quarter was US$1.4 billion, a year-on-year decrease of 38%. %, which is the fastest year-on-year decline among several items listed.
"Not only for Intel, but you can look at the entire semiconductor industry in the past two years. Everyone's profits have been reduced by about 20%-30%. But recently, we have seen that the economy is gradually recovering, and there is also a lot of demand. Whether it is in the cloud or with the further deployment and use of AI technology, especially the demand for AI, it is not only the demand for GPU and network, but also the demand for CPU is constantly increasing. There will also be some demand at the edge. For Intel, what we do is to adapt to market conditions and changes." Sachin Katti said.
- END -
If you are also optimistic about the edge computing market and eager to change this market, then we sincerely invite you to participate in the 8th Global Edge Computing Conference held in Shanghai at the end of October! Collaborate with the best edge computing companies in the Yangtze River Delta to develop new trends in the edge computing industry!
The Shenzhen station of the Global Edge Computing Conference ended successfully. The link in the picture above is the video playback and PPT download link. Thank you for your support to the edge computing community. Welcome to the 2023 Global Edge Computing Conference Shanghai Station!