Raimondo prevents China from obtaining computing power and restricts Chinese companies from accessin

Mondo Finance Updated on 2024-02-11

Raimondo prevents China from obtaining computing power and restricts Chinese companies from accessing U.S. data centers!

Biden's administration does not allow China to use American technology to develop AI.

U.S. Secretary of Commerce Raimondo confirmed two important options in an interview: one is to prohibit Nvidia from "exporting advanced AI chips"; Second, it has imposed strict controls on the entry of Chinese enterprises and individuals into the information center in the United States.

In fact, the first two options are used to prevent us from gaining mathematical abilities.

Computing power has become a hot topic, and many companies have listed computing power as their priority development direction, such as Nvidia, Huawei, etc.

What exactly is math all about? Is there a fart in math? In fact, most people don't have enough knowledge to give precise and comprehensive answers.

The most intuitive meaning of mathematics is operation, that is, the ability to process data and data.

When we talk about CPU and GPU, it's actually a narrow mathematical concept. Because only the CPU and GPU cannot complete complex algorithm calculations, they must be combined with memory chips and software to complete the calculations.

The NVIDIA H100, for example, claims to have 800 transistors and 2,000 TFLOPS, but it requires 94 GB of memory and relies on the NVIDIA CUDA development platform.

Among them, generalized computing includes CPU and GPU (narrow computing), memory and hard disk (storage), operating system, application (algorithm) and other layers.

Our country has to do computing, not only graphics processors, but also memory chips, operating systems, and databases.

Otherwise, once we move into graphics processing units, the United States will re-impose a blockade on our memory and operating systems.

What is the significance of arithmetic?

Former U.S. Secretary of State Henry Kissinger once said, "If you master oil, you will master the world," and now the math problem has been replaced by "oil in the 21st century."

Mathematics can be as revolutionary as the steam engine of the eighteenth century, the electricity of the nineteenth century, and the computer of the twentieth century.

Our mobile phones, computers, tablets, and every day we work need to be calculated. Online shopping, Didi and takeaway are all counted by numbers. Cloud computing, big data, blockchain, and other cutting-edge technologies can all be applied to arithmetic.

Medical research, space programs, nuclear testing, and other very sophisticated tasks require a lot of computing power.

In short, all information technology is based on mathematics.

The mathematics of our nation is in **?

China is the world's second-largest computing power, growing at a rate of nearly 30% every year.

It is estimated that by the end of 2022, China's overall computing capacity will exceed 180 megabytes, the storage capacity will exceed 1,000 megabytes, and the communication delay between nodes in China will be less than 20 ms, supporting the development potential of the industry to exceed 18 trillion.

We have a population of more than 1.4 billion, but our numeracy is still very poor, and we are a country with moderate numeracy skills.

In addition, we also have mathematical difficulties.

For example, the increase in computing power cannot keep up with the increasing computing power; Domestic AI chips have insufficient single-block computing capabilities and cannot support large-scale models with a scale of 100 billion yuan for learning.

Americans have seen the power of mathematics, and our weakness in mathematics has increased the constraints on us.

1) NVIDIA's output of high-end AI chips has been restricted.

NVIDIA is a leader in the GUP industry, with 82% of the world's AI accelerators and 95% of AI training equipment, and is a leader in the AI industry.

According to IDC statistics, the current market share of AI accelerators in China has reached 1.09 million, with Nvidia accounting for 85%, Huawei accounting for 10% and 2%, and Cambrian and Suwon Technology accounting for 1%.

It can be seen that just by controlling NVIDIA's AI chips, it is equivalent to choking the throat of China's highly dependent Chinese AI industry.

To this end, the U.S. Department of Commerce proposes the following rules: (1) chips with a computing capacity greater than or equal to 4,800 tops will be subject to export controls; (2) For data streams greater than or equal to 600 Gbs, they need to be throttled.

Originally, NVIDIA's A100 and H100 were restricted, but Nvidia was unwilling to let go of China's opportunity and entered the Chinese market and launched A800 and H800.

However, the U.S. Department of Commerce used before this product was put on the market"or"Instead"with"and changed the number of nodes from 4800 TOPS and 600 GB s to 4800 TOPS and 600 GBS.

In a word, it is a big change. After doing so, the A800 and H800 will not be available for sale, and Nvidia will have to "castrate" them again.

Samsung Electronics plans to launch the HGXH20, L20PCLE, L2PCLE and others in China, all of which are based on NVIDIA's CUDA architecture for training, inference, and edge computing.

The H20's equivalent computing power is only 296 Tflops, which is about 15% compared with the H100, which is not even as good as Huawei's Ascend910B, and there is a huge gap in computing power.

What's more, after the H20 cluster, its computing power cannot reach the megabytes of big data such as Huawei's Pangu, Tencent's Mixed Yuan, and Alibaba's Qianjie.

In this context, in the face of the downturn in H20, both Tencent and Alibaba have made it clear that there is no need, and we are preparing to place orders from domestic chip manufacturers.

But the United States hit back.

Limit the output of high-end semiconductor devices.

If you want to make a chip, you must have a complete set of semiconductor devices, and the United States, Japan and the Netherlands occupy 91 of the world5% of semiconductor products, if these three giants are missing, it is equivalent to a broken leg.

To this end, the United States, Japan, the Netherlands and other countries have imposed restrictions on their products, requiring them to obtain export licenses for logic chips below 14 nm, DRAM below 16 nm, and NAND flash memory below 128 nm.

Although the three countries did not specify that they were targeting Chinese companies, anyone with a discerning eye could see it.

As a result, Chinese companies have purchased ASML lithography units months before the ban was imposed to keep domestic chips running.

In October, the United States and Canada were 700 million and 2200 million and 1.1 billion, up 10 times respectively compared to the same period last year.

From the United States, Japan, and the Netherlands, there will be nearly $40 billion in imports by 2023.

However, EUV lithography machines are not one of them. At present, the best EUV lithography equipment in the world is ASML's 2100 i, which can achieve mass production of 7 nm process, but its performance is not guaranteed.

The NVIDIA H100 and H200 are based on 4 nm process technology, while the next generation will use 3 nm process.

The development of lithography process needs to be developed by ourselves, which will become an inevitable bottleneck for China's AI chips to catch up.

We're still making progress in lithography, and the U.S. is already hyping it up.

Data centers in the United States are restricted.

Raimondo, of the U.S. Department of Commerce, said in an interview that if China bypasses the problem of AI and uses the U.S. "cloud" to cultivate larger models, then why should we bother to restrain Nvidia? Why is this road closed?

At the same time, Biden's cabinet will also propose that U.S. cloud computing companies decide whether foreign companies can access a U.S. data center to train their AI models.

The U.S. has the largest number of data centers in the world, rising to 2,670 by the end of 2022 with 153 in Dallas, Los Angeles and 137 in the Bay Area.

Like data centers in Northern Virginia, data centers in Dallas, Silicon Valley, Phoenix, and many more, like Virginia, Dallas, Silicon Valley, Phoenix, and many more.

Because of the unique system in the United States, many private companies are actively building their own data centers, such as Amazon, Microsoft, Google, and they are able to provide cloud technology.

According to a research report, the world's top 10 cloud computing companies are: Amazon, Microsoft, Google, Alibaba, Oracle, IBM, Tencent, OVH, Digital Ocean, and Linode.

Of the 10 largest manufacturers in the world, seven are from the United States, and only two are Chinese: Alibaba and Tencent.

If the U.S. prohibits Chinese companies from using the business of U.S. cloud computing providers, there will be few opportunities for Chinese companies in overseas markets, as are those that are not even as good as Alibaba or Tencent in their home markets.

In the short term, the United States' miscalculation this time has had a great impact on us, such as high-end AI chips cannot be purchased, domestic chips also have certain restrictions, and overseas cloud computing systems are temporarily unusable.

The lack of mathematics will be a new problem that restricts our development.

Therefore, there is still a big gap in our computing power, and we still need to continue to work hard to establish a corresponding software system based on China's independent research and core hardware.

In addition, when the computing power reaches a certain level, we must also carry out rational planning to maximize the computing efficiency and eliminate the waste of computing resources.

Each of us can devote our computer resources to work and research, not just enjoy 3 seconds of fun.

I'm Science & Technology, if you think it's good, like it!

Related Pages