At the AI computing power conference, Lin Yuan, CEO of Qingyun Technology, shared what capabilities Qingyun Technology will provide to support surging computing power at the advent of the new era of AI.
As you can clearly see, the proportion of AIGC investment in the digital IT infrastructure of enterprises is increasing rapidly year by year. As the expectation of AIGC investment continues to accelerate, the demand for AI computing power from enterprise customers is also growing rapidly year by year. It's an unstoppable AI trend.
AI and large-scale model production tools have brought more revolution. In the past 10 years, computing power has been more focused on supporting mobile Internet and enterprise digital applications. At the core of enterprise applications, CPUs are still dominant and more latency-sensitive. At that time, many data centers and CPU-based computing centers were deployed in economically developed coastal cities.
In the past 10 years, our applications have mainly consisted of traditional applications and cloud-native applications. AI and large models are a watershed moment, and in the next 10 years, applications, requirements, and their underlying architecture will also change dramatically. In the future, artificial intelligence and digital transformation should be the mainstay, and the new business form will be different from the past, which will be GPU-based, and will involve training and inference, which will be both cost-sensitive and latency-sensitive, and the deployment of the entire computing center is scattered.
Computing power needs to support traditional applications, support cloud-native applications, and support AI applications that will grow rapidly in the future. With the trend of AI, more things have changed, and there are a lot of problems that IT has to solve. Focusing on the current stage, what are the key challenges?
AI includes computing power, algorithms, and data. At this stage, it is the construction stage of computing power, and intelligent computing centers have sprung up like mushrooms after a rain, and the construction process is in full swing in various places. After the construction of these intelligent computing centers, how to operate them has become a key challenge for everyone. Expand to see:
The investment in the Intelligent Computing Center is very huge. Since the investment is huge, how to monetize the original investment and how to make a profit has become a more stressful matter. It depends on the costs versus the benefits. Cost is the operational efficiency of the intelligent computing center, and revenue is whether the platform capabilities of the intelligent computing center can support the business of various customers.
Medium- and long-term operations are more specialized and complex issues. At present, the construction participants of the intelligent computing center are very diverse, including local enterprises, integrators, central enterprises, AI companies, GPU manufacturers, etc., so that the construction is scattered, and the operational capabilities of investors and builders are uneven.
From the perspective of the most critical factor - supporting applications, including traditional applications, AI applications, and cloud-native applications, the support of application diversification determines the efficiency of operations and the carrying capacity of future services.
There are a lot of challenging issues that need to be addressed during the operational phase. For example, how to consider design and planning? We must not be a single computing power, we must have a unified scheduling platform that can support multiple computing power such as supercomputing, intelligent computing, and cloud computing, and can support diversified applications, which also needs to be supported by an efficient underlying architecture, and must be an efficient, operable, and operationally manageable platform.
To meet this big challenge faced by the investors and builders of the intelligent computing center, the answer of Qingyun Technology is a software product - Qingyun AI intelligent computing platform, which is not only a key tool for the operators of the intelligent computing center, but also can help the operator establish a closed loop from construction to operation.
Let me briefly list a few differentiating highlights:
1. Diversified computing power. Because it needs to support various applications, including enterprise applications, scientific research applications, smart city applications, traditional applications, etc., the Qingyun AI computing power scheduling platform supports diverse computing power.
2. Unified operation and maintenance and operation under the same set of platforms. The Tsing Cloud AI intelligent computing platform is backward compatible and interconnected with different hardware, different CPUs, and different GPUs to support various applications. Not only that, the Qingyun AI intelligent computing platform also has the ability of ultra-large-scale deployment and operation, and has very successful operation experience and successful cases.
This is the first software product we have released this time, the Qingyun AI Intelligent Computing Platform, which is aimed at the operators of the Intelligent Computing Center.
From another perspective, let's see what kind of AI computing power they need from the perspective of end customers, which is our second officially released service this time - Qingyun AI Computing Cloud Service. The needs of the end customer are complex:
1.From training to data cleansing to inference.
2.From traditional to cloud-native to AI.
3.Support for application life cycle such as development, testing, launch, and operation.
4.Cost-sensitive and latency-sensitive, private and public.
From the perspective of the best business in the ecological chain, everyone is different, but they need to form a synergy with each other.
1.There are investors, builders, and operators throughout the cycle of the intelligent computing center.
2.The construction of the computer room, the hardware layer, the chip layer, the scheduling layer, the driver layer, the model layer, and the application layer are all different.
3.The first-class merchants at each level are very professional, and they all need to invest in long-term time accumulation, as well as huge manpower and capital investment.
Therefore, Tsing Cloud believes that AI computing power cloud services should be an open ecological alliance, and there are three key words here:
The first keyword: ecology. Because what the customer needs is a complete solution, the ability of one family is not enough, it needs the ability of an ecosystem to do aggregation.
The second keyword: openness. Because each participant is very focused and professional, the participants need each other and are open to each other.
The third keyword: alliances. This is very important, the entire AI computing cloud should be a very long-term thing, and here our ecological partners should be an alliance, and they must be like-minded and win-win in the long run.
To sum up in one sentence: Qingyun hopes to build an "AI Cloud" with our ecological alliance through an open, mature, and operable AI computing power scheduling platform in a variety of ways, such as self-operation, joint venture, and support for third-party operations.