NVIDIA fully capitalizes on the GPU dividend, is GeForce 256 a great starting point?

In August 2024, the domestically produced game "Black Myth: Wukong" was officially released. With the complexity and detail of game graphics achieving a major enhancement, this signifies the full arrival of an era. The GeForce 256 not only propelled the advancement of the gaming industry but also extended this concept to other fields. Although Nvidia had not introduced revolutionary technology at the time, GeForce 256 indeed laid the foundation for the application of GPUs in areas such as scientific computing and financial analysis. At the 2024 Yunqi Conference, Alibaba Group CEO and Chairman and CEO of Alibaba Intelligent Cloud, Wu Yongming, emphasized: Wu Yongming pointed out that the AI era will usher in a new computing model. It is predicted that in 2024, more than 50% of the new computing power demand in the market will be driven by AI, a trend that will only intensify in the future. It is not difficult to see that even Alibaba Cloud's AI cloud computing system is mainly relying on the powerful driving capabilities of GPUs. However, concentrating a large number of GPU resources in the cloud and combining them with local devices' CPUs and NPUs for hybrid computing is also an effective strategy. First of all, this move can effectively alleviate the bottleneck of computing power. While GPUs are irreplaceable in AI computing, the high costs of purchasing and maintaining GPUs have always been a barrier for individual developers and ordinary consumers. By centrally providing GPU resources in the cloud, users can rent as needed and flexibly allocate computing power based on actual tasks. Moreover, many terminal devices are restricted by size and power consumption and cannot be equipped with high-performance GPUs. By processing heavy AI computations in the cloud and distributing the results locally for simple task execution, devices can remain lightweight and low power consumption. The effective collaboration between local CPUs and NPUs can maximize the computing power of cloud GPUs: the local NPU can quickly handle latency-sensitive tasks, such as speech recognition and real-time image processing, while complex model training and inference can be entrusted to cloud GPUs. This model can significantly reduce computing response time and greatly enhance user experience. At the current stage, Nvidia still holds a significant advantage in the AI computing power market, but from a long-term perspective, the highly competitive AI market inevitably presents greater challenges for Nvidia. Markets such as mobile PCs and smart terminals may gradually be eroded by other manufacturers, and the rapid development of AMD's MI300 series AI graphics cards in the server market also needs attention. Competition brings not threats but drives technological innovation and progress. As Wu Yongming said, a new ecosystem centered around AI is gradually unfolding before us.