Text | Hearing the sound of the green tree
Edit | Hearing the sound of the green tree
Sources: All of the information stated in this article is based on reliable sources, and is detailed at the end of the article
When the U.S. takes a technological lead in a certain area, it often uses its technological advantage to block the development of other countries in the industry, and the same is true in the field of AI.
After the United States made a breakthrough in the field of AI, the first thing it did was to stop the mainland's progress in the field of AI, and the United States' method was to "restrict the mainland's access to computing power".
Stop the development of AI in the mainland
The development of AI is inseparable from computing power, and excellent AI software is built with huge computing power. The thing that can provide a large amount of convenient computing power for AI is the GPU, which is what we usually call a computing power chip.
The reason why the United States has the best AI technology in the world is because the United States has Nvidia, which makes computing chips.
Although the mainland is inferior to the United States in terms of AI, the mainland's technology companies are making rapid progress and are the most powerful competitors of AI in the United States, so the United States will never let the mainland progress in the field of AI.
The U.S. crackdown is simple: start at the source and prevent the mainland from acquiring high-end GPUs. After all, if the mainland's AI wants to develop rapidly, it is also inseparable from computing power.
As early as 2022, the United States banned Nvidia from exporting A100 high-end computing chips to the mainland. At the same time, the United States uses the A100 chip as the standard, stipulating that the transmission speed of computing power chips exported to the mainland shall not exceed 600GB per second.
However, Nvidia did not want to give up the mainland, a big buyer, so it tailored the A800 computing power chip for the mainland.
In order to meet the export standards of the United States, Nvidia adjusted the transfer speed of the A800 computing power chip to 400GB, which is only two-thirds of the A100 model.
For high-tech companies in the mainland, although the A800 chip is not as good as the A100, it is enough, and a large number of chips are still exported to the mainland. Seeing this situation, the United States once again issued new regulations, vowing to cut off the mainland's AI path.
In October 2023, the United States once again issued new control regulations, banning the export of five types of computing power chips in one go: NVIDIA A100, A800, H100, H800, and L40S.
This regulation of the United States is really hurtful to others and hurts itself, the technology companies in the mainland cannot get computing chips, and the American company Nvidia cannot make money, and the two sides have reached a stalemate.
In addition, AI computers equipped with H800, even if their performance is far inferior to H100, are still being coaxed in the international market, with a price of more than 2.3 million yuan per unit and a single card of about 400,000 yuan.
Even so, technology companies in various countries are worried that there is no graphics card available, and they have rushed to buy this chip, and for a while, the AI computer equipped with H800 has no price.
Mainland technology companies (such as Baidu, Tencent, ByteDance, etc.), even if they are willing to buy the "castrated" H800, they can't get the spot, so they can only place orders and wait slowly. These technology companies have made achievements in the field of AI, and if they do not have access to computing chips, the research and development progress of AI will be greatly delayed.
Nvidia didn't want to suffer this loss, and launched the H20 computing power chip for the mainland, but the computing power of this chip is really not good, only 20% of the H100, in addition to the lower price, the advantages of this chip are very limited.
As a result, mainland technology companies are not interested in H20 and are only trying to buy some to find new ways to break the game.
The U.S. export ban has also hit Nvidia hard. Previously, mainland buyers would buy nearly a quarter of NVIDIA's computing chips, and after the restriction, Nvidia would reduce its income a lot.
Nvidia founder Jensen Huang also said that China is both a component source for U.S. chips and a third of the market for U.S. technology companies. In the long run, the export ban will affect the revenue and development of computing chips in the United States.
In any case, mainland technology companies can't stop the research and development of AI, and if they can't get NVIDIA's computing power chips, Baidu, ByteDance and other companies have looked at Huawei's Ascend AI chips.
Blessing and misfortune: Huawei's opportunity has come
Unlike other technology companies, Huawei has been sanctioned by the United States many times, so Huawei attaches great importance to self-research and development, and Huawei has its own products for all important high-tech components, including computing chips.
Huawei's more well-known computing chips are the Ascend 910 and Ascend 310, which fill the gap in this field in the mainland.
Especially the Huawei Ascend 910B, this chip is much stronger than the H20 castrated by NVIDIA in terms of performance, and it can almost be compared with NVIDIA's A100!
Although the Ascend 910B computing power chip is not very high-end and cannot be compared with high-end chips in the United States, it can definitely meet the needs of various technology companies.
Previously, many companies did not choose the Ascend 910B because of ecological and adaptation issues. After all, the Ascend 910B is too different from the NVIDIA GPU, and technology companies will have to pay a lot of money if they use the Ascend 910B.
Huawei is also well aware of this problem, so Huawei continues to improve the research and development of Ascend chips. Today, there are 1.5 million developers of the Ascend series of chips and 25 AI computing centers across the country.
These developers have continuously improved the adaptability of Ascend chips, helping to expand Ascend chips across the country, ushering in the spring of Huawei's AI chips.
Without the U.S. ban, it would be difficult for tech companies to choose Huawei. After all, even if Huawei's computing chips can be used, they can't be compared with NVIDIA's top computing chips.
Moreover, Nvidia chips have been sold worldwide for many years, and the software and algorithms of various technology companies are based on Nvidia's research and development.
The U.S. ban has given Huawei hope that many domestic companies will have to adopt Huawei's chips, and Huawei's AI chip division will also be able to gain funding and experience to narrow the gap with Nvidia.
More importantly, Huawei can take this opportunity to cultivate its own chip ecology in China and cultivate its own users.
The race for chips is a marathon
In the Securities Commission's filing, Nvidia has listed Huawei as its biggest competitor in the chip sector. It can be seen that Huawei's chips are paired with the huge market in the mainland, which has made Nvidia feel threatened.
Like Huawei, Nvidia sees top companies such as Intel, Qualcomm, and Microsoft, but these companies are highly dependent on foreign markets and do not have a vast domestic market like Huawei.
At present, the world's AI chips are divided into three categories: the best chips come from established chip companies such as Nvidia and AMD, but both of these are American companies and need to comply with the US ban.
The second-class AI chips come from technology giants such as Huawei and Google, and these companies also have to develop AI chips in order to build their own general models, but the chips of these technology companies are worse than those of Nvidia.
The third-tier companies are some smaller chip companies, such as Cambrian, which have limited production capacity and design chips that are mainly "small and beautiful".
Today, 80% of the world's high-end computing chip market is occupied by NVIDIA, and other companies can only share the remaining share. Huawei still has a long way to go before it can get out of the "shadow" of Nvidia.
The mainland does not need to worry about the development of AI chips for the time being. NVIDIA's achievements today are the result of decades of deep research by this company, and Huawei, as a rising star, is already very good to have today's achievements.
Moreover, the competition of AI is not a race against time, but a marathon race. The mainland not only has to compete with AI technology, but also with electricity, materials and other conditions, and there are still many places where the mainland can do it.
Moreover, technology companies on the mainland are also actively looking for ways to break the situation. Previously, Baidu and other companies borrowed cloud computing centers in the United States to train domestic big data models with the help of the powerful computing power of the United States.
This model of "borrowing chickens to lay eggs" is very effective, because the cloud computing center in the United States has a large number of high-end chips, and the computing power is quite powerful, which is better than the computing power chips purchased by the mainland. However, such an approach was quickly rejected, and the U.S. government did not allow foreign companies to borrow computing power.
This example also shows that mainland technology companies are not out of nowhere, and there are many ways to obtain computing power, but it costs more and is more difficult.
epilogue
The ban of the United States has indeed prevented the mainland from obtaining computing power and slowed down the development of AI in the mainland. However, doing so will also cause Nvidia to lose the world's largest chip market, lose the opportunity to grow revenue and industrial development, and in the long run, technology companies in both countries will suffer greater losses.
Fortunately, the mainland also has Huawei's Ascend chips. Although Ascend chips are not as good as high-end computing chips in the United States, they can also support the development of AI in China.
If there is no US ban, Huawei wants to "break through" under the shadow of Nvidia, and I don't know when it will have to wait. The blockade by the United States has given Huawei the opportunity to cultivate its ecosystem and develop higher-grade chips.
In time, the mainland's computing chips and AI technology will inevitably be able to make great progress, even if it can't catch up with the United States in all aspects, it can make breakthroughs in several key projects.
In fact, AI is a technology that benefits the whole world, and it is impossible for any one country to monopolize all the technologies and occupy the entire market.
Therefore, instead of being wary of each other, it is better to cooperate for a win-win situation, China and the United States lead the world in AI technology, if they can communicate and cooperate with each other, AI applicable to all mankind can appear earlier, and the world of the United States and the United States and the world will not be far away.
Past Highlights:
AI showdown between China and the United States: 54% of talents and 73% of capital in the United States, and GPU chips are even more monopolized
Resources:
Hatps://baijiyahaao.baidu.com/s?id=1793951917261609922&dfr=spider&for=pc
http://news.youth.cn/gj/202308/t20230824_14741271.hatama
Hatps://vv.chinnav.com/m/gj/2023/06-29/10033689.
hatps://baijiyahaao.baidu.com/s?id=1748905252655384148&dfr=spider&for=pc
Disclaimer: The description process and pictures of the article are all from the Internet, and this article aims to advocate positive social energy and no vulgar and other bad guidance. If it involves copyright or character infringement issues, please contact us in time, and we will delete the content as soon as possible!
#MCN首发激励计划#