Can domestic GPU chips replace NVIDIA and AMD?
As we all know, the use of chips and GPUs in China is becoming increasingly restricted.
In 2022, the United States will completely ban the sale of high-end graphics processors such as A100 and H100, among which desktop graphics processors such as A800, H800, L40, L40S, and even RTX 4090 will be completely banned from 2023.
The U.S. doesn't want China to overtake the U.S. in graphics processing units because it has developed a high-performance GPU.
Some people may say, can domestic graphics processors break through the encirclement? Can it replace AMD and NVIDIA?
In fact, there are two main types of GPUs nowadays, one is the general GPU, which is used for drawing, and we can also think of it as a tool to speed up the game.
According to the data, Intel, AMD, and NVIDIA have a share of more than 98% in the GPU rendering field, while in China, this proportion is less than 2%.
Although the market share is still very low, it is no worse than AMD and NVIDIA in terms of performance. For example, the latest release of Cogobuy City"Fenghua No. 1"The 4 K GPU, which performs a bit better than the NVIDIA GTX980, uses 12 nm process technology.
In addition, the MTTS80 graphics card, which has a separate Moore thread, is also based on NVIDIA's NVIDIA process, which can achieve 144 Floating-point calculations for TFLOPS.
In addition, companies such as Jingjia, although they have also launched several domestic rendering graphics processors, but because they were developed relatively late, they are used less, but if you want to replace the top, it is the top.
In addition, NVIDIA A100, H100, A800, H800, etc. are also a new type of AI graphics card, which can adapt to new application scenarios such as AI modeling and analysis.
In the field of AI image processing, NVIDIA accounts for 97%, AMD accounts for 2%, and other manufacturers together account for less than 1%.
This part of the hunting is very efficient, just like the NVIDIA H100, steady and steady.
Now, many domestic companies have released some AI graphics processors, such as Huawei's Ascend AI chip, Ascend 910 B, and so on. This AI chip can be compared to the Nvidia A100, but it is still too far behind the H100.
In addition, NVIDIA also has a rich CUDA environment, which can abstract various interfaces and models into different interfaces, and many of the current AI learning and inference are built in the CUDA environment.
Since the GPU used in the home cannot be connected to the CUDA computing platform, to replace the AI chip, the original CPU operating environment has to be changed, and the previous learning with CUDA has to be restarted.
Therefore, domestic GPUs may catch up in the short term, but from an ecological point of view, it is still a long process, and the process of being replaced is also quite long.
But, as Nvidai himself said, if they don't sell GPUs to China, then China will develop them themselves, and there's no doubt about that, because there's no one else to sell, so they have to recharge themselves.
To put it simply, the current top-of-the-line GPUs cannot be replaced, they can only be replaced step by step, this is a kind of compulsion, once changed, Nvidia, AMD and other brands will withdraw from the Chinese market forever, this thing is ** with the green logo, if you want to completely replace it, you have to start all over again.