Can domestic GPU chips replace NVIDIA and AMD?
As we all know, the United States has now imposed stricter controls on China's integrated circuits, including GPUs.
In 2022, the United States banned high-end graphics processors such as A100 and H100, and in 2023, A800, H800, L40, L40S, etc., even the RTX 4090 was banned.
The U.S. doesn't want China to overtake the U.S. in terms of GPUs because of its strong GPU computing power.
So many students are wondering, can the domestic GPU keep up? Can it replace AMD and AMD?
In fact, there are two types of GPU chips nowadays, one is the general GPU chip, which is used for rendering, and can also be said to be used for games and image display.
Although most of the products are in research and development, they do not occupy a large proportion, according to the data, Intel, AMD and Nvidia, occupy 98% of the world's GPU, while China's GPU combined, is less than 2%.
Although the market share is small, but in terms of performance, they are not inferior to AMD, AMD and other companies, take Innosilicon as an example, they released a 4K graphics processor called "Fenghua No. 1", using a 12 nm process, the performance is slightly better than NVIDIA's GTX980.
There's also Moore Threads' MTTS80, which is specifically designed for gamers and can support 14With 4 tflops of floating-point hashrate, it can compete with NVIDIA.
Jingjiawei and other companies have also launched their own graphics processors, but because they were developed too late and not so fast, they do not occupy much of the market, but if they are used to replace them, it will be better.
Like NVIDIA's A100, H100, A800, H800, etc., are all designed for the requirements of computing power in the fields of artificial intelligence modeling and computing.
Now, in the field of AI graphics cards, NVIDIA occupies more than 97% of the market, while AMD occupies 2% of the market, and other companies may not even have 1%.
And NVIDIA's H100 is the best choice.
Now domestic AI processors have begun to be developed, such as Huawei's Feiteng AI, Feiteng 910B, etc., although in some aspects, it can be compared with NVIDIA's A100, but compared with H100, it is still a big difference.
In addition, NVIDIA also has a very good CUDA environment, which can be said to be an interface, a model, and a lot of AI training and inference are now based on CUDA.
If you want to replace the AI chip, you must change the operating system, and the model made with CUDA before has to be redesigned, which is very complicated.
Therefore, although domestic GPU chips can keep up in terms of speed, they still have a big gap in terms of software, and it will take some time to replace them.
However, as Nvidai said, they didn't bring GPUs to China, but let China produce them themselves, which is undeniable, because no other manufacturer wants to, so they have to have their own products.
In short, the current high-end GPU, can not be replaced, can only be replaced step by step, this is the situation is forced, if we replace AMD, AMD and other companies, will forever withdraw from the Chinese market, this product, has been with our industry, if replaced, it is forever eliminated.