Shulou(Shulou.com)11/24 Report--

With the eight big models successively passing the "interim measures for the Management of generative artificial Intelligence Services" for the record, Wen Xin took the lead in opening up to the whole society, AIGC will really usher in a period of actual landing and flowering in China, and the "hundred model war" will become more and more intense.

Behind the popularity of AIGC driven by ChatGPT, the demand for artificial intelligence computing is also skyrocketing. Public data show that GPT-4 's model has nearly 1.8 trillion parameters and 13 trillion training data, and the cost of one training is as high as 63 million US dollars. The dependence of large models on computing power is obvious. In addition to increasing the supply of computing resources, it is also very important to optimize the utilization rate of computing power and lower the threshold for use.

There is no doubt that the realization of intelligent and collaborative scheduling of computing power, algorithms and applications, and the transformation of computing power into productivity more efficiently, has become the key to winning the "computing" era.

As the core information infrastructure enterprise covered by the whole industry chain, dawning's support and experience for large-scale model and large-scale computing power is far earlier than 2023.

In recent years, based on the in-depth insight and forward-looking prediction of the industry, dawning has built all kinds of computing centers nationwide, including the Computing Center, the Cloud Center and the Integrated big data Center, and has mastered rich, diverse and versatile computing resources. Through intelligent scheduling and resource management, computing resources will be connected to the cloud, and take the innovative computing service platform as the export to provide easily accessible computing power and applications to users in different industries and regions.

Up to now, dawning has participated in the training, fine-tuning and reasoning of many kinds of large models, such as "Wudao 2.0", "Zidong Taichu", "Wen Xin Yi Yan", "Tongyi Thousand questions" and so on, to provide strong arithmetic support. It has completed the adaptation and incubation of more than 30 mainstream models at home and abroad, including GPT series, LLaMA series, GLM series and so on. Under the same conditions, the large model training efficiency and training stability have been significantly improved.

In addition, in view of the challenges brought by large model training, dawning works closely with ecological partners to work closely with Baidu, Ali, Zhiyuan Research Institute, Zhisu AI, Fudan University, Zhijiang Laboratory, Jettong Huasheng and other industry leading enterprises and scientific research institutes to fully mobilize the efforts of politics, industry, learning, research and utilization to create large model all-in-one machines and large model storage solutions for different application scenarios. Provide full-stack AI capabilities from underlying computing power, frameworks, algorithms, applications, etc., to meet the needs of multiple scenarios, and accelerate the exploration and innovation of large models and the landing of industrialization.

At the same time, due to the huge demand for computing power of the large model, the problem of energy consumption is becoming more and more serious. Since the beginning of 2011, dawning has devoted itself to related exploration, developed the leading immersion phase change liquid cooling technology and cold plate cooling technology, built a full-stack green computing solution including computing and storage, and has successfully realized commercial deployment. Compared with the traditional air-cooled mode, the core components of AI calculation can be cooled by 20-30 °C at the same power, and the energy consumption for computing equipment can be reduced by 30% under the same performance. At present, more than 58% of liquid-cooled data centers in China have chosen dawning technology.

At every stage of the evolution of the information industry, there will be a mainstream technology application direction. In the "ever-changing" environment, only grasping the forward-looking trend and insisting on building core competitiveness is the "constant" rule to deal with the change. Whether it is a big model or the next new "tuyere", dawn has long been waiting.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.