Raimondo restricts China's computing power, and enterprises access U.S. data centers
It is obvious that Biden is preventing China from using American technology to develop artificial intelligence.
U.S. Secretary of Commerce Raimondo confirmed the news in an interview in which he made two recommendations: First, restrictions'NVIDIA AI chips exported to foreign countries; The second is to impose restrictions on Chinese companies or individuals from accessing U.S. data centers'。
Actually, both Scenario 1 and Scenario 2 are designed to prevent us from gaining computing power.
The term computing power has become a hot word, often heard in ** work reports, and many companies also make computing power the focus of their development, such as Nvidia and Huawei.
What exactly is computing power? What is computing power used for? In fact, most people are unable to answer these questions.
Literally, arithmetic refers to the ability to process information and data.
What is commonly referred to as CPUs and GPUs is actually computing power in the strict sense of the word. Because CPUs and GPUs alone can't provide enough computing power.
For example, NVIDIA's H100 has 80 billion transistors and 2,000 TFLOPS of computing power, but requires 94 GB of memory, which also requires NVIDIA's CUDA technology.
In general, computing includes the CPU, GPU, memory, hard disk, operating system, and applications.
When it comes to computing power, we're not limited to GPUs, but memory chips, operating systems, and databases.
Otherwise, even if we attack the graphics processor, the United States will suppress us in terms of memory and operating systems.
What is the value of arithmetic?
Former U.S. Secretary of State Henry Kissinger once said:"The domination of oil is the domination of the world"In the 21st century, computers are becoming synonymous with oil.
Computing power can produce powerful artificial intelligence that will revolutionize humanity, just as the steam engine did in the 18th century, electricity in the 19th century, and computers in the 20th century.
Our phones, computers, and tablets all need computing power. Whether it's online shopping or on behalf of others, we need computing power. Cloud computing, big data, blockchain, and other cutting-edge technologies are all examples of computing power applications.
Medical research, space programs, and nuclear testing all rely on powerful processing power.
In conclusion, everything related to computer science is inseparable from arithmetic.
How is your math?
China is the second largest country in the world in terms of total computing power, with an average annual growth rate of nearly 30%.
By the end of 2022, China's total computing capacity will reach 180 trillion operations per second, the storage capacity will exceed 1,000 BS (1 trillion GB), the national core network latency will be reduced to 20 milliseconds, and the industrial scale of core computing power will reach 1.8 trillion yuan.
With a population of 1.4 billion, our country lags far behind other countries in terms of computing power.
In addition, there are some problems with computing power.
For example, the growth in computing power cannot keep up with the growing demand for computing power; The computing power of a single national AI chip card is too low to support large-scale, multi-billion dollar model training.
The United States saw the huge potential of our computing power, and also saw the shortcomings of our country in this field, so it began to suppress our computing power.
1.Restrict the export of Nvidia's high-end artificial intelligence chips.
NVIDIA is a far leader in GUP, with an 82% market share in the AI space.
According to IDC **, by 2022, the shipment volume of Chinese intelligent acceleration board market will reach 1.09 million, of which Nvidia accounts for 85%, Huawei accounts for 10%, accounts for 2%, Cambrian accounts for 1%, and Suwon Technology accounts for 1%.
Obviously, Nvidia is very dependent on the Chinese market, and if Nvidia cannot export AI chips to China, China's AI industry will be strangled in the cradle.
As a result, the U.S. Department of Commerce determines that 1) a chip has more than 4,800 TOPS of computing power and 2) a transfer rate greater than 600 GB s, then the chip falls under the export control category.
Nvidia's A100 and H100 are export-controlled products, but Nvidia didn't want to give up the Chinese market, so it launched the A800 and H800 for the Chinese market.
However, just some time before the release of this chip, the US Department of Commerce will be the original"4800 tops "Instead"600 gb/s"and will"with"Instead"or"。
The difference between one word and the difference is worlds apart. This made the A800 and H800 unmarketable, and NVIDIA had to neutralize them further.
The HGXH20, L20PCLE, and L2PCLE are three chips for the Chinese market, all of which are based on the NVIDIA CUDA architecture for training, inference, and edge computing, respectively.
With only 296 TFLOPs, the H20 has only 15% of the H100's computing power, which is far less than the Huawei Ascend 910B.
In addition, for companies with billions of dollars of computing power, such as Huawei's Pangu, Tencent's Hybrid, and Alibaba's Q&A, the H20's computing power is far from enough.
Both Tencent and Ali expressed their willingness to place orders with domestic chip manufacturers.
But the United States is here again.
Export restrictions on high-end semiconductor equipment.
Chip production is based on semiconductor equipment, while the United States, Japan, and the Netherlands control 915% of semiconductor equipment, therefore, to put it mildly, as long as these three countries are not there, any chipmaker should be at a standstill.
As a result, the United States, Japan, and the Netherlands began restricting the export of semiconductor equipment such as sub-14nm logic chips, sub-16nm DRAM, and 128-layer NAND flash memory, which must be approved before export.
While none of the three countries have explicitly stated that they want to deal with Chinese companies, it would be expected with a keen eye.
As a result, Chinese companies began buying lithography equipment from ASML months before the ban on domestic chips went into effect.
The value of imports in October was 6700 million yuan, 8200 million yuan, up to 1.1 billion yuan in December, a 10-fold increase over the same period last year.
This year, nearly 40 billion yuan worth of equipment was imported from the United States, Japan and the Netherlands.
However, this does not include ultraviolet lithography; The best lithography machine in the world today is ASML's Type 2100 I, which is capable of producing products at 7 nanometers, but the yield is difficult to guarantee.
Meanwhile, NVIDIA's H100 and H200 use a 4nm process, and future products will use a 3nm process.
Therefore, in order to realize the national artificial intelligence chip, lithography technology must be developed independently, which is an insurmountable gap.
We were working on a lithography machine at the time, and it turned out that there was something else in the United States.
Restrictions on access to U.S. data centers.
U.S. Commerce Secretary Raimondo said in an interview that if China can solve the problem of artificial intelligence chips and use the U.S. cloud system to train large models, then what can stop Nvidia from doing so? Therefore, we must close the gap.
In addition, Biden** suggested that it be up to U.S. cloud computing companies to decide whether foreign entities can access U.S. data centers used to train AI models.
By 2022, the number of data centers in the U.S. will reach 2,670, with 153 in Dallas and 137 in Los Angeles and the Bay Area.
For example, data centers in Virginia, Dallas, Silicon Valley, and Phoenix.
Thanks to the unique system in the United States, many private companies are also setting up their own data centers, such as Amazon, Microsoft, and Google, that can provide cloud computing services.
According to a study some time ago, the top 10 cloud computing providers in the world are Amazon, Microsoft, Google, Alibaba, Oracle, IBM, Tencent, OVH, Digital Ocean, and Lynyrd.
Seven of the top 10 producers are U.S. companies, while only two in China are Alibaba and Tencent.
If the U.S. blocks Chinese companies from using the services of U.S. cloud service providers, then Chinese companies will have no choice abroad, as other companies are not as strong as Alibaba and Tencent.
In the short term, the performance of the United States in terms of computing power has a great impact on us, we cannot buy advanced artificial intelligence chips, we cannot independently develop chips, and we cannot provide cloud computing services abroad.
At a time when we are in dire need of computing power, the lack of computing power will become a new problem that hinders our rapid development.
Therefore, in terms of computing power, we still have a long way to go, and we must rely on our efforts to solve our core technologies and build good hardware and software systems.
At the same time, it is necessary to reasonably schedule existing computing power to improve computing efficiency and avoid waste.
As ordinary people, we should put our limited computing power into work and study, instead of just enjoying three seconds of happiness.
I'm tech minting, thanks, I love it!