Nvidia releases the "strongest chip" and how it affects the artificial intelligence track in China and the United States
Nvidia releases the "strongest chip", the image is Nvidia's data image/IC photo
On March 18th local time, the highly anticipated Nvidia GPU Technology Conference (GTC) opened, and Nvidia CEO Huang Renxun released a series of highly anticipated AI (Artificial Intelligence) new products, including the latest generation GPU architecture "Blackwell", GPU chip B200 based on this architecture, GB200 superchip based on B200 GPU, and GB200 next-generation computing unit "GB200 NVL72 based on GB200.
In his speech, Huang Renxun used phrases such as "the engine driving a new round of industrial revolution", "the factory of AI industrial revolution", "accelerated computing has reached a critical point, and universal computing is outdated" to describe the products released this time. But what gives the most intuitive impression to the outside world is that Nvidia has launched a product with stronger computing power. Some media claim that Nvidia has released the strongest AI chip in history.
AI becomes smarter faster
According to Huang Renxun, the new product will improve the performance of large model inference workloads by 30 times, while significantly reducing costs and energy consumption. This means that training a model with 1.8 trillion parameters previously required 8000 Hopper GPUs (chips) and a power consumption of 15 megawatts. Now, the same task can be completed with only 2000 Blackwell GPUs and a power consumption of only 4 megawatts.
Simply put, Nvidia launched a more powerful GPU chip at the GTC conference, and then built various large-scale AI supercomputing clusters through chip and chip connectivity technology. In the future, under the leadership of Nvidia products, computing power performance will be greatly improved, expanding AI to trillions of parameters.
Correspondingly, the speed at which AI becomes smarter will be faster, and the entire industry ecosystem will also undergo changes due to the surge in computing power. The AI industry will witness a moment of change.
Some comments refer to Nvidia's new GPU chip B200 as a "nuclear bomb," which is a slightly exaggerated comparison. The B200 is strong, but it still represents an improvement in computing power, which cannot be compared to the power of nuclear bomb invention to change the world pattern, at least so far.
This can actually be seen from the fluctuation of Nvidia's real-time stock price, where the stock price rose only 0.70% on the day, indicating that the B200 did not exceed market expectations too much.
But the significant improvement in computing power is still exciting. In the era of AI, computing power is one of the three major elements of AI (algorithms, data, and computing power), and it is actually the biggest bottleneck in the development of AI.
Previously, there was a belief that the first half of AI was driven by computing power. With more and more developers of AI and large models, computing resources are becoming increasingly scarce, and computing power driven AI has reached a bottleneck.
As a result, Nvidia, which shoulders the responsibility of leading the way in breaking bottlenecks, has received high expectations from the market, and its stock price has almost doubled since the beginning of the year. The launch of B200 has at least narrowed the computing bottleneck a lot, although it did not exceed expectations, it has also fulfilled market expectations.
OpenAI company logo taken in 2023/Xinhua News Agency
AI competition is a marathon run
From the other side of the ocean, the launch of B200 has actually added pressure to China's AI industry. Because this means that in the competition of AI and computing power, it is necessary for China to further accelerate, otherwise the gap will continue to widen.
From a macro perspective, if artificial intelligence competition is viewed as a big game, it is beneficial for Chinese companies to temporarily encounter bottlenecks in computing power improvement, as we can catch up through comprehensive factors such as data and applications.
In recent years, China's AI industry has sprung up as a new force. Relying on the world-class Internet industry and huge data resources, China and the United States are both known as the world's top AI companies. But with the emergence of ChatGPT, Chinese AI industry suddenly felt a "mountain of pressure".
Upon closer observation, it can be observed that behind ChatGPT lies OpenAI, which is a computing infrastructure provider like Nvidia. The world-class competition of AI has actually returned to infrastructure competition. China urgently needs to have leading enterprises in this area to address the bottleneck problem at the infrastructure level.
China is also facing a bottleneck in the field of chip manufacturing, but this is mainly due to historical reasons, as China's chip industry started too late and catching up takes time.
The rise of Nvidia is different because it has been around for less than a decade. Prior to this, it was just a technology company specializing in the gaming market, but in a short period of time, it became the world's strongest artificial intelligence enterprise, and there are many things worth learning from.
For example, in terms of focus, although Nvidia has evolved from providing gaming graphics cards for PCs to providing chips for artificial intelligence enterprises, fundamentally, it has not changed the track of its 31 year entrepreneurship. Whether it's users playing games, mining, or running large models, it always adheres to the GPU as the core for research and development, but constantly expands its application scenarios. This is actually a kind of long-term ism, adhering to the main business and investing in scientific research and innovation, ultimately waiting for its own trend.
For example, without emotions, one cannot persist. Nowadays, the world sees Nvidia and OpenAI being interdependent, but in 2016, when OpenAI was still unknown, Nvidia CEO Huang Renxun personally drove and delivered the world's first AI supercomputer as a gift to OpenAI.
At that time, there were over 100 companies waiting in line for goods, but Huang Renxun still donated the first product to OpenAI. He signed on the machine: In order to calculate and contribute to the future of humanity, I donated the world's first DGX-1.
It is worth mentioning that B200 is strong, but Chinese companies do not need to be too nervous. China currently does not have its own NVIDIA, but there will eventually be one. For example, Huawei's artificial intelligence chips have become quite powerful.
No one can stop the rise of China's AI, just like when asked about Nvidia's first listing of Huawei as a major competitor in multiple categories such as AI chips, Foreign Ministry spokesperson Mao Ning said: "A small courtyard and a high wall" cannot stop the pace of China's innovative development.
China is one of the world's largest semiconductor markets and also one of the world's largest AI application markets. At the same time, computing power has been listed as a key factor in promoting economic development by the country, and the construction of the "East West Computing" project is in full swing.
In his speech at Stanford University, Huang Renxun made it clear that "the end of AI lies in photovoltaics and energy storage! We cannot just think about computing power. If we only consider computers, we will need to burn 14 Earths of energy. Super AI will become the bottomless pit for electricity demand."
China leads the world in the fields of new energy and renewable energy generation. Looking at this, the world competition about AI is undoubtedly a marathon, and it is far from the time when the outcome is determined.
