Welcome new opportunities in artificial intelligence: the unlimited potential and continuous evolution of Qingyun AI

Feixiang News (Wei Deling/Text) "Han wants troops, the more the better" must be a popular allusion. The Marquis of Huaiyin used this to describe his ability to lead troops, and of course Gaozu followed him. The "good general". The advantages of "the more the better" and "good generals" are becoming a necessary capability in the industry under the wave of AI.

Analysts predict that the artificial intelligence market will usher in a boom that will last more than ten years. In enterprise information technology, the proportion of investment in artificial intelligence will become higher and higher. Artificial intelligence will stimulate demand 10 to 100 times that of the past decade, and the corresponding demand for computing power will also show the same growth. The guarantee to meet this demand is "multiple" choices, because the supercomputing center hardware resources called behind different computing power requirements are not the same. As the saying goes, “more choices, more laughter.” The Jinan Center, a national supercomputing center that has achieved profitability, is a good example

More choices, more laughter

In terms of current common AI use case requirements, many traditional scientific computing applications carried out by universities require HPC to complete tasks such as simulation, simulation, ocean current prediction, genetic testing, etc.; for traditional government cloud business, Traditional CRM/ERP enterprise digital applications are traditional cloud computing needs, which require the use of CPU resources; currently popular smart city applications involving road recognition and license plate recognition require the use of GPU capabilities. The diverse needs of business exactly determine the need for diverse computing power.

According to reports, just three years ago, Jinan’s National Supercomputing Center already had 1000P computing power and more than 300PB storage capacity, becoming one of the largest computing centers in Asia at the time. The center includes high-performance computing, traditional cloud computing based on CPU, and intelligent computing based on GPU. In the field of intelligent computing, the center takes into account domestic needs and introduces some domestic GPUs, making it a supercomputing center with both multiple computing capabilities and heterogeneous computing capabilities

"The benefits generated by investing in building a platform depend on the platform's capabilities, because the stronger the platform's capabilities, the more types of businesses it can support customers. The more customers there are, the clearer the profit model." Participated in serving the National Supermarket three years ago Lin Yuan, president of Qingyun Technology, which operates the Jinan Center, explained to the media why heterogeneous computing and diverse computing power are needed.

On the contrary, if you do not embrace multiple computing power and just build a supercomputing center that only supports a certain kind of hardware computing power as its core, you will often face embarrassment in subsequent actual operations. For example, a major domestic manufacturer once built a computing center in a certain place, but because it only supported a certain GPU, when application requirements came, incompatibility problems occurred, resulting in a narrow customer base

However, how to schedule 1000P of computing power and 300PB of storage, integrate diversity and heterogeneity, and enable the supercomputing center to achieve efficient operations, just like Han Xinbing, the more the better, instead of falling into a management dilemma, the same is A problem that needs to be faced. As a participant in the construction of the National Supercomputing Jinan Center, Qingyun Technology provides it with the ability to lead the overall situation. This requires mentioning the new product recently released by the company-AI computing power scheduling platform.

Qingyun AI, the more the merrier

We have achieved success in the practice of the National Supercomputing Center in Jinan, which proves the capabilities and achievements of the Qingyun platform. Lin Yuan is very confident in this product. Qingyun AI computing power scheduling platform is an important tool for computing center operators and can establish a closed loop from construction to operation. In the case of the National Supercomputing Jinan Center, we not only successfully implemented the practice three years ago, but also helped the center achieve good operations and profitability

Qingyun AI computing power scheduling platform has the same management capabilities as "Han Xinbing, the more the better". It can uniformly manage multiple resources such as GPU computing power, HPC computing power, multiple storage systems, model resources and data resources, and realize computing Automated management of power platforms. In addition, the platform can also distribute resources according to the needs of different industries, and has distributed scheduling and management capabilities. It can automatically allocate and manage computing resources, greatly shortening task execution time, improving work efficiency, and allowing customers to focus on business innovation and applications. Development

The management and operation of the platform is also a piece of cake. Through the unified operation and operation and maintenance management platform, it can realize the post-operation of operational services in multiple service scenarios, standardize and efficient operation and maintenance, and help users achieve refined operations. Qingyun will provide platform administrators with a visual large-screen management interface that can easily browse thousands of computing resources and storage across the country. Taking the National Supercomputing Center in Jinan as an example, the back-end actually only requires a team of 10-20 people to implement software operation and maintenance and solve various customer problems.

Qingyun Technology uses an operation and maintenance platform to enable pricing, discounts, promotions and other operations to be completed by just clicking buttons on the page. Miao Hui, product manager of Qingyun Technology, said that compared with daily processes such as traditional cloud computing application forms, Qingyun's AI computing power scheduling platform has obvious advantages in operation and maintenance efficiency

At the same time, Qingyun AI computing power scheduling platform can also help customers solve problems such as network switching speed, environment construction, and multi-service integration.

In addition to greatly improving the management capabilities of platform operation and maintenance personnel, Qingyun can also further improve the development efficiency of AI implementation for platform users. For algorithm engineers, the development host provided by Qingyun has built-in development environments and IDE environments. Engineers can directly upload Python project files, write code, debug and run online, and immediately find the required training cluster. After the inference is completed, algorithm tuning can also be implemented, and the code can be continuously optimized through the computing cluster and model.

In addition, Qingyun also provides an online training platform that integrates high-performance computing and GPU cards. In a dedicated environment, Qingyun provides an option to apply for a GPU server to build a cluster online, and all networks and environments can be generated with one click. At the same time, Qingyun will also integrate according to commonly used models and gadgets in the industry

The container inference platform supports one-click deployment of high-performance Kubernetes clusters. During the inference process, if you encounter a performance bottleneck, you can implement load balancing and elastic scaling

Qingyun also provides a model warehouse so that customers can immediately deploy their own models and call their own services through the model market. At the same time, the model can be fine-tuned or obtained online through the cloud platform with one click.

In general, Qingyun AI computing power scheduling platform manages AI infrastructure in a manner similar to managing local resources, providing diversified computing power scheduling and intelligent computing power scheduling to ensure that computing power can be put into use quickly

Open the ecosystem and let customers become “good generals”

As a Qingyun AI computing power scheduling platform with the capability of "the more, the better", the second question we face is whether there is an opportunity for users to also have the strength of "good generals". In fact, in the future AI era, due to huge investment, it may be difficult for a single enterprise to achieve comprehensive coverage, because the cost of each aspect will be higher than in the past decade. For example, a large model may require a lot of manpower, computing power, money and time costs, and the same is true for CPU and IDC construction. Therefore, Qingyun is working with partners in different fields to achieve the integration of capabilities

We have a bold idea that a giant will emerge in the AI ​​field. However, we believe that this giant should not be an independent company but a small ecosystem. Lin Yuanlong introduced the positioning of Qingyun AI computing power cloud service, which is to operate together through an open ecological alliance

This idea is mainly based on three considerations. First, the ecosystem will be able to meet customers' needs for complete solutions. Second, each professional participant has each other's needs and maintains development with each other. Ultimately, like-minded people will form a long-term cooperation. Winning alliance.

From the perspective of customer needs, the complete solution to customer needs includes computer room, computing power, scheduling platform, model, model driver, Model Service, and executable applications. Each layer, from computing power, models, services to application scenarios, requires strong professionalism.

Qingyun AI computing power ecosystem has been built around the ecological sharing of AI computing power infrastructure, the resource integration of AI computing power large models, and the ecological integration of AI data resources, thereby better helping enterprises achieve real business value. During the media interview, Lin Yuan classified it into a technical ecology and a business ecology. The technical ecology includes the GPU and model ecology, and the business ecology includes the investment and construction of computing centers, AI end customers, etc.

At present, Qingyun has realized the integration of the ecosystem from adaptation, MaaS, co-construction to final implementation, covering multiple levels of cooperation from major well-known chip manufacturers, model developers, cloud data center providers, etc. For customers of Qingyun AI Computing Cloud, they are like AI pioneers who can mobilize a large number of famous generals to achieve the effect of "good generals".

Lin Yuan said that at the arrival of each new era, people are on the same starting line, and the new era may require some new cooperation modes and gameplay. He believes that the arrival of the artificial intelligence era may bring about new changes in the landscape

The success story after "The more the merrier" goes without saying. Nowadays, AI is being regarded as another new opportunity after cloud services. Qingyun AI, which has "can do it, done it, and succeeded", has also happened to show its performance. With the confidence of "the more, the merrier", the next exciting chapter is beginning.

The above is the detailed content of Welcome new opportunities in artificial intelligence: the unlimited potential and continuous evolution of Qingyun AI. For more information, please follow other related articles on the PHP Chinese website!