Whether domestic GPU chips can replace NVIDIA and AMD is a mystery to be solved.
The U.S. is now tightening restrictions on Chinese chips, and the same is true for GPUs.
The United States will ban the use of high-performance graphics processor chips such as A100 and H100 in 2022, and the use of A800, H800, L40, L40S, and even RTX 4090 for desktops in 2023.
The U.S. doesn't want China to overtake the U.S. in GPUs because of its GPU lead.
Some people may say, will our GPUs rise? Can it be an alternative to AMD and NVIDIA?
In fact, now GPU chips can be divided into two types, one is the commonly used GPU chip, which is used for rendering, and can also be understood as the ordinary graphics card used for game acceleration and graphics display.
According to the data, Intel, AMD, and NVIDIA occupy 98% of the entire rendering graphics processor market, while domestic manufacturers have less than 2%.
Fenghua No. 1"It is a high-performance 4 K graphics card graphics processor, it uses 12nm process, core dynamics and other technologies, and its performance is a little better than NVIDIA's GTX980.
In addition, Moore Threads has also launched the MTTS80 discrete graphics card, which can support up to 144 TFLOPS of floating-point calculations and support for NVIDIA gaming.
Companies like Jingjiawei, although they have also launched some domestic rendering-level graphics cards, but because of the relatively short development time, the market share is not very large, but in terms of performance, they are likely to surpass these two brands.
NVIDIA's A100, H100, A800, H800, etc., are designed for the needs of new computing capabilities such as artificial intelligence modeling and reasoning.
In the field of AI graphics cards, Nvidia has a market share of more than 97%, while AMD has only 2%, and the market share of other manufacturers may be less than 1%.
Just like Nvidia's H100, this pure high-performance product has thoroughly outperformed its competitors.
Now, there are Huawei's Ascend AI chips, as well as Ascend 910 B, these artificial intelligence chips, in some performance, are stronger than NVIDIA's A100, but if you want to compare with H100, it is still 108,000 miles behind.
In addition, NVIDIA also has a strong CUDA ecosystem, which can be understood as a variety of interfaces, models, etc.
The original GPU can't use the CUDA ecosystem, so if the AI chip is replaced, you have to change the ecology, and the graphics that were originally trained with CUDA have to start all over again, which is very troublesome.
Therefore, although domestic GPU chips can quickly keep up with in terms of performance, they are still a big gap in terms of ecology, and it is impossible to replace them for a while.
However, as Nvidai himself said, if the GPU is not to China, then China will develop the GPU on its own, there is no doubt about it, because no one wants to, and even if our GPU can't do it, we will do it.
In short, high-end GPUs are impossible to be replaced, and it is impossible to be replaced, however, it will be replaced step by step, which is a forced choice, if we replace it, HTC, AMD and other brands will lose the Chinese market forever, because it is related to the survival of mankind.