The first leader in AI computing power, the annual report increased by 1520, and the main force incr

Mondo Finance Updated on 2024-03-08

The domestic computing power market is quietly changing ......To solve the problem, we have to start from the background, the supply and demand of the domestic computing power market are changing directionally.

The first is the supply of resources, that is, the market that provides computing power, has changed. Since last year, there has been a situation in the domestic market where "N cards are hard to find". To this end, Nvidia has taken turns to launch special versions of GPUs, which have repeatedly shrunk in computing power and power, but they are still limited.

The HGX H20 and two new GPUs, which have only recently been reported, have also been revealed to be delayed until February or March next year. A series of shrinking actions in the supply of computing power have made the existing NVIDIA series graphics cards in the domestic market further scarce, and it is the norm for computing power to increase by 50% or even 100% within a month.

According to the "Economic Information Daily", due to the continuous shortage of computing resources, domestic computing service companies such as Winner Technology have planned to increase the computing service fees of high-performance computing power servers embedded with NVIDIA A100 by 100% in mid-November.

At the same time, the rapid growth of computing power demand from domestic Internet manufacturers due to large models has exacerbated this tense situation. First of all, the number of large models in China has surged, and statistics in October show that 238 large models have been released in China. A survey of large language models shows that the 65 billion parameter large model LLAMA was trained on 2048 80G A100 for 21 days. The 70 billion large model LLAMA 2 also used 2,000 pieces of 80G A100 for training.

However, the demand for AI computing power will continue**. According to OpenAI's estimates, since 2012, the computing power required for AI model training has doubled every 3 to 4 months, and the computing power required to train AI models has increased by up to 10 times every year. Under the influence of multiple catalysis, AI computing power is expected to become a new market thread that relays Huawei's industrial chain.

To this end, after an in-depth review, I have excavated 3 potential leaders of "AI + computing power concept", especially the last one, remember to save the collection, and the new partners will pay attention! The first: Radio and Television Express.

In terms of computing power deployment, the company has completed the integration of the computing center and the Yuntong cloud platform, built an AI data center computer room, a GPU computing power sharing platform and a high-performance cloud computing platform, and realized the unified planning of GPU computing power and cloud computing resources, as well as the closed-loop management of on-demand allocation.

The second: VenustechThe company's self-developed Pangu AI platform serves as the AI base for security operations, and based on the natural language model trained by the special corpus for security operations, it can accurately understand various questions about related tasks in security operations and use professional security operation terminology to answer them for users.

The third: the most optimistic one at the moment, optimistic for the following reasons:

In the direction of computing power + AI, the company continues to increase R&D investment in heterogeneous parallel architecture, immersion liquid phase change cooling technology, high-speed interconnection technology, advanced computing management, operation and maintenance, operation service platform software, and industry software ecological construction

At present, the stock price is in an upward trend after being fully sufficient, and the main force is obviously involved.

The main low-level heavy position is absorbed, and the recent net inflow exceeds 8 billion.

The medium-term idea, the MACD monthly level golden cross, the expected space is 60%+.

In order not to disturb the layout of the main force, it is still the bottom ** adjustment stage, waiting for time, soaring, what about friends who need to know? Not much to say here, it's ready to start, interested friends on the homepage send the number three six to learn about AI computing power

Related Pages