AMD announced that it will host a live event with the theme of "Advancing AI" in the near future, which will introduce the MI300 series of next-generation data center GPU accelerators. According to reports, AMD's data center GPU revenue is expected to reach $400 million in the fourth quarter of this year and exceed $2 billion in 2024. The MI300 may become the company's fastest product to sell more than $1 billion, and Microsoft, Google, Amazon, Meta and other companies are interested in buying.
At this event, AMD also upgraded its AI software ecosystem to enhance the performance and functionality of its ROCM platform. ROCM is integrated into the mainstream PyTorch and TensorFlow ecosystems. ROCM is also known as "CUDA-like", Haiguang DCU (General GPU) is compatible with the mainstream ecosystem - open-source ROCM, supports mainstream deep learning Xi frameworks such as TensorFlow, PyTorch and PaddlePaddle, and adapts to mainstream application software. Haiguang DCU ROCM has an excellent ecology, and Shensuan-2 has been commercialized and can be used for large model training. The research and development of Shensuan No. 3 went smoothly.
In addition, on December 5, 2023, People** published an article entitled "Let Computing Power Be Used", emphasizing the importance of allowing users to call computing resources anytime and anywhere, so as to lower the threshold for use and expand application scenarios. The article points out that computing power has become a basic resource for the benefit of the public, and the relevant policies for the construction of a national integrated computing power network are expected to continue to be introduced. The construction of the national computing network will focus on the construction of public computing power platforms, the construction of computing power dispatching centers, the participation of small and medium-sized enterprises, and the integration of computing power and electricity.
In terms of investment, it is recommended to pay attention to the following types of companies: Haiguang Information, Cambrian, Yunsai Zhilian, High-tech Development, and Sichuan Changhong related to the construction of domestic computing power and public computing power service platformSugon, Sitech, Zhonghe Technology, Parallel Technology, and Qingyun Technology related to computing power scheduling and operation;Cloud vendors participate in the construction of computing networks related to Shensangda, Tongniu Information, Capital**, and Youkede;and Yakang shares related to the integration of computing and electricity.
However, we must also be aware of the associated risks. First of all, there may be a risk that the introduction of policies related to computing power may be less than expected. Second, there may be a risk that the demand for computing power will not be as expected. In addition, there may be a risk that the production capacity of chips will not be as expected. All of these factors may have an impact on the performance of the companies involved.
In general, with the rapid development and widespread application of artificial intelligence, the demand for computing power is also increasing. In the process, the innovation and progress of AMD and other companies in computing hardware and software have attracted much attention. At the same time, enterprises also need more cooperation and support in the construction and application of computing power.
Note: The data and opinions in this article are provided by the Oriental Computer team and are for reference only.
A shares